Merge branch 'main' into luzk/flow-token-usage

- Protect _aggregated_usage_metrics with a lock so concurrent
  LLMCallCompletedEvent handlers can't race the read-modify-write
  inside add_usage_metrics, and so usage_metrics snapshots are
  consistent.
- Wire the usage aggregation listener into resume_async so LLM
  calls during outcome collapsing and downstream crews continue
  to roll up into flow.usage_metrics after a paused-then-resumed
  kickoff. Restores current_flow_id to the original kickoff's
  match id when none is set, and detaches in finally.
- Guard against reentrant kickoff on the same Flow instance:
  only the outer kickoff captures _flow_match_id, resets the
  accumulator, and owns the listener lifecycle. Inner reentrant
  calls pass through and no longer wipe outer state or detach
  the shared handler.
- Rename test_snapshot_is_immutable to
  test_usage_metrics_returns_independent_copy to reflect that
  the property returns a copy of a (still-mutable) UsageMetrics.
- Extend test_handler_is_unregistered_after_kickoff to also
  cover the failure path, confirming the handler is removed
  when kickoff raises.

.github/workflows/vulnerability-scan.yml

@@ -64,6 +64,7 @@ jobs:
             --ignore-vuln PYSEC-2025-197 \
             --ignore-vuln PYSEC-2025-210 \
             --ignore-vuln PYSEC-2026-139 \
+            --ignore-vuln GHSA-rrmf-rvhw-rf47 \
             --ignore-vuln PYSEC-2025-211 \
             --ignore-vuln PYSEC-2025-212 \
             --ignore-vuln PYSEC-2025-213 \
@@ -81,6 +82,7 @@ jobs:
         #   PYSEC-2025-183      - pyjwt 2.12.1: disputed weak-encryption claim; key length is application-chosen
         #   PYSEC-2025-189..197 - torch 2.11.0: memory-corruption/DoS in functions only reachable via untrusted models; no fix available
         #   PYSEC-2025-210, PYSEC-2026-139 - torch 2.11.0: profiler/deserialization issues; no fix available
+        #   GHSA-rrmf-rvhw-rf47 - torch 2.11.0 (CVE-2025-3000, alias of PYSEC-2025-194): memory corruption in torch.jit.script, CVSS 1.9, local-only; affected <=2.12.0, no fix available. pip-audit reports it under the GHSA id so the PYSEC ignore above does not catch it.
         #   PYSEC-2025-211..218 - transformers 5.5.4: deserialization/code injection via malicious model checkpoints; no fix available
         #   GHSA-f4j7-r4q5-qw2c - chromadb 1.1.1 (CVE-2026-45829): pre-auth RCE via /api/v2/tenants/{tenant}/databases/{db}/collections when trust_remote_code=true.
         #                         Advisory: vulnerable >=1.0.0,<=1.5.9, firstPatchedVersion=none. We only use chromadb.PersistentClient (lib/crewai/src/crewai/rag/chromadb/factory.py)

.pre-commit-config.yaml

@@ -47,6 +47,7 @@ repos:
           --ignore-vuln PYSEC-2025-197
           --ignore-vuln PYSEC-2025-210
           --ignore-vuln PYSEC-2026-139
+          --ignore-vuln GHSA-rrmf-rvhw-rf47
           --ignore-vuln PYSEC-2025-211
           --ignore-vuln PYSEC-2025-212
           --ignore-vuln PYSEC-2025-213

docs/ar/changelog.mdx

@@ -4,6 +4,106 @@ description: "تحديثات المنتج والتحسينات وإصلاحات
 icon: "clock"
 mode: "wide"
 ---
+<Update label="11 يونيو 2026">
+  ## v1.14.7
+
+  [عرض الإصدار على GitHub](https://github.com/crewAIInc/crewAI/releases/tag/1.14.7)
+
+  ## ما الذي تغير
+
+  ### الميزات
+  - إضافة واجهات خلفية افتراضية قابلة للتوصيل للذاكرة، والمعرفة، وrag، وflow.
+  - عرض السبب الحقيقي للإنهاء، ومعلمات العينة، وresponse.id في أحداث LLM.
+  - تصنيف مشغلات DSL كزخارف واعية للمسار.
+  - إضافة واجهة برمجة تطبيقات الدردشة لتدفقات المحادثة.
+  - جعل واجهة القفل قابلة للتجاوز.
+  - بناء FlowDefinition من بيانات التعريف الخاصة بـ Flow DSL.
+  - إضافة مزود LLM من Snowflake Cortex الأصلي.
+  - إضافة دعم لملفات الوكلاء المدربين من crew.
+
+  ### إصلاحات الأخطاء
+  - إصلاح نقطة التحقق لإعادة بناء BaseLLM مخصص كـ LLM ملموس عند الاستعادة.
+  - تقييد الاستعادة على علامة لمنع اللقطات الحية من إعادة التشغيل كاستئناف.
+  - تحديد حالة وقت التشغيل لكل تشغيل للحد من النمو وعزل التشغيل المتزامن.
+  - إصلاح إعدادات التتبع على crewai-login.
+  - احترام suppress_flow_events لأحداث تنفيذ الطريقة.
+  - استعادة [project.scripts] في حزمة crewai لتثبيت أداة uv.
+  - حل مشكلات CVE الخاصة بـ pip-audit لـ aiohttp وdocling وdocling-core.
+  - إصلاح إدخال الملفات الذي لا يعمل بشكل موثوق.
+  - إصلاح تاريخ نتائج أدوات Snowflake Claude غير المكتملة.
+
+  ### الوثائق
+  - تحديث سجل التغييرات والإصدار لـ v1.14.7.
+  - تحديث وثائق جامع OpenTelemetry.
+  - تحديث دليل NVIDIA Nemotron LLM.
+  - إضافة دليل تكامل Databricks.
+  - إضافة دليل تكامل Snowflake.
+
+  ### الأداء
+  - تحسين سرعة استيراد crewai من خلال تحميل مستندات docling بشكل كسول.
+
+  ### إعادة الهيكلة
+  - تبسيط تقييم شروط التدفق ليكون بلا حالة لكل حدث.
+  - فصل منطق المحادثة عن وقت التشغيل وإضافة تعريف المحادثة.
+  - تقسيم `flow.py` إلى DSL، وتعريف، ووقت تشغيل.
+
+  ## المساهمون
+
+  @Luzk, @alex-clawd, @devin-ai-integration[bot], @greysonlalonde, @gvieira, @jessemiller, @lorenzejay, @lucasgomide, @mattatcha, @vinibrsl
+
+</Update>
+
+<Update label="10 يونيو 2026">
+  ## v1.14.7rc2
+
+  [عرض الإصدار على GitHub](https://github.com/crewAIInc/crewAI/releases/tag/1.14.7rc2)
+
+  ## ما الذي تغير
+
+  ### إصلاحات الأخطاء
+  - استعادة البوابة على علامة لمنع اللقطات الحية من إعادة التشغيل كاستئناف
+
+  ### الوثائق
+  - تحديث سجل التغييرات والإصدار لـ v1.14.7rc1
+
+  ## المساهمون
+
+  @greysonlalonde
+
+</Update>
+
+<Update label="10 يونيو 2026">
+  ## v1.14.7rc1
+
+  [عرض الإصدار على GitHub](https://github.com/crewAIInc/crewAI/releases/tag/1.14.7rc1)
+
+  ## ما الذي تغير
+
+  ### الميزات
+  - إضافة `reset_runtime_state` لإطلاق حالة الحافلة المتراكمة
+  - التعامل مع دعم كل من الموجهات المخصصة
+  - فصل منطق المحادثة عن وقت التشغيل وإضافة `conversational_definition`
+
+  ### إصلاحات الأخطاء
+  - إصلاح نطاق حالة وقت التشغيل لكل تشغيل للحد من النمو وعزل التشغيلات المتزامنة
+  - إصلاح إعدادات القياس عن بُعد على `crewai-login`
+  - إصلاح احترام `suppress_flow_events` لفعاليات تنفيذ الأساليب
+
+  ### الوثائق
+  - تحديث صور OpenTelemetry
+  - تحديث الوثائق لتعكس الحالة الجديدة لجمع بيانات OpenTelemetry
+  - تحديث سجل التغييرات والإصدار لـ v1.14.7a4
+
+  ### إعادة الهيكلة
+  - تبسيط تقييم شرط التدفق ليكون بلا حالة لكل حدث
+  - تحسين دورة توجيه المحادثة مع تقليل مسار واحد
+
+  ## المساهمون
+
+  @greysonlalonde, @lorenzejay, @lucasgomide, @vinibrsl
+
+</Update>
+
 <Update label="9 يونيو 2026">
   ## v1.14.7a4
 

docs/ar/concepts/flows.mdx

@@ -226,6 +226,48 @@ counter=2 message='Hello from first_method - updated by second_method'
 من خلال ضمان إعادة مخرجات الدالة الأخيرة وتوفير الوصول إلى الحالة، تجعل تدفقات CrewAI من السهل دمج نتائج سير عمل الذكاء الاصطناعي في التطبيقات أو الأنظمة الأكبر،
 مع الحفاظ على الوصول إلى الحالة طوال تنفيذ التدفق.
 
+## مقاييس استخدام التدفق
+
+بعد اكتمال تنفيذ التدفق، يمكنك الوصول إلى الخاصية `usage_metrics` لعرض إجمالي استخدام التوكنات عبر **كل استدعاء لنموذج اللغة** يتم خلال التشغيل — بما في ذلك الاستدعاءات من كل فريق (Crew) ينظمه التدفق، والاستدعاءات داخل أدوات الـ Agents، والاستدعاءات المباشرة لـ `LLM.call(...)` من دوال التدفق. هذا هو المكافئ على جانب الـ SDK للإجماليات المعروضة في واجهة CrewAI Enterprise.
+
+```python Code
+from crewai import LLM
+from crewai.flow.flow import Flow, listen, start
+
+class UsageMetricsFlow(Flow):
+    @start()
+    def run_first_crew(self):
+        self.state.first_result = FirstCrew().crew().kickoff()
+
+    @listen(run_first_crew)
+    def call_llm_directly(self):
+        # استدعاء مباشر لنموذج اللغة — يُحسب أيضًا ضمن flow.usage_metrics
+        llm = LLM(model="openai/gpt-4o-mini")
+        self.state.summary = llm.call("لخّص النقاط الرئيسية.")
+
+    @listen(call_llm_directly)
+    def run_second_crew(self):
+        self.state.second_result = SecondCrew().crew().kickoff()
+
+flow = UsageMetricsFlow()
+flow.kickoff()
+
+print(flow.usage_metrics)
+# UsageMetrics(total_tokens=8579, prompt_tokens=6210, completion_tokens=2369,
+#              cached_prompt_tokens=0, reasoning_tokens=0,
+#              cache_creation_tokens=0, successful_requests=5)
+```
+
+<Note>
+  `flow.usage_metrics` **ليست** نفس `flow.kickoff().token_usage`. هذه الأخيرة
+  ترجع فقط `CrewOutput.token_usage` لـ **آخر** دالة `@listen` أعادت
+  `CrewOutput`، مما يعني أنها تعكس فقط الفريق الأخير وتتجاهل الفرق السابقة
+  وكذلك أي استدعاءات مباشرة لـ `LLM.call(...)`. استخدم `flow.usage_metrics`
+  كلما احتجت إلى الإجمالي **الكامل** للتوكنات لتنفيذ التدفق.
+</Note>
+
+كل حقل في [`UsageMetrics`](https://github.com/crewAIInc/crewAI/blob/main/lib/crewai/src/crewai/types/usage_metrics.py) المُعاد هو مجموع جميع استدعاءات نموذج اللغة التي حدثت خلال استدعاء واحد لـ `flow.kickoff()`. تتم إعادة تعيين العدادات عند الاستدعاء التالي لـ `kickoff()` (وفي كل تكرار من `kickoff_for_each`)، لذلك لن تتكرر العدّات عبر التشغيلات المتتالية. يمكن قراءة هذه الخاصية بأمان في أي وقت بعد اكتمال `kickoff()`؛ قراءتها أثناء التنفيذ تُرجع المجموع الجزئي المتراكم حتى تلك اللحظة.
+
 ## إدارة حالة التدفق
 
 إدارة الحالة بفعالية أمر بالغ الأهمية لبناء سير عمل ذكاء اصطناعي موثوق وقابل للصيانة. توفر تدفقات CrewAI آليات قوية لإدارة الحالة غير المهيكلة والمهيكلة،

docs/ar/enterprise/guides/capture_telemetry_logs.mdx

@@ -24,15 +24,39 @@ mode: "wide"
 
 1. في CrewAI AMP، انتقل إلى **Settings** > **OpenTelemetry Collectors**.
 2. انقر على **Add Collector**.
-3. اختر نوع التكامل — **OpenTelemetry Traces** أو **OpenTelemetry Logs**.
-4. هيّئ الاتصال:
-   - **Endpoint** — نقطة نهاية OTLP لمجمّعك (مثل `https://otel-collector.example.com:4317`).
-   - **Service Name** — اسم لتعريف هذه الخدمة في منصة المراقبة.
-   - **Custom Headers** *(اختياري)* — أضف رؤوس المصادقة أو التوجيه كأزواج مفتاح-قيمة.
-   - **Certificate** *(اختياري)* — قدم شهادة TLS إذا كان مجمّعك يتطلبها.
-5. انقر على **Save**.
+3. اختر تكاملاً:
+   - **OpenTelemetry Traces** و**OpenTelemetry Logs** — صدّر إلى أي مجمّع أو واجهة خلفية متوافقة مع OTLP.
+   - **Datadog** — أرسل التتبعات مباشرة إلى استقبال OTLP الخاص بـ Datadog، دون الحاجة إلى مجمّع منفصل أو Datadog Agent.
+4. هيّئ الاتصال. تعتمد الحقول على التكامل الذي اخترته:
 
-<Frame>![تهيئة مجمّع OpenTelemetry](/images/crewai-otel-collector-config.png)</Frame>
+<Tabs>
+  <Tab title="OpenTelemetry Traces / Logs">
+    إن **OpenTelemetry Traces** و**OpenTelemetry Logs** تكاملان منفصلان يتشاركان نفس الحقول — اختر التكامل المطابق للإشارة التي تريد تصديرها.
+
+    - **Endpoint** — نقطة نهاية OTLP لمجمّعك (مثل `https://otel-collector.example.com:4317`).
+    - **Service Name** — اسم لتعريف هذه الخدمة في منصة المراقبة.
+    - **Custom Headers** *(اختياري)* — أضف رؤوس المصادقة أو التوجيه كأزواج مفتاح-قيمة.
+    - **Certificate** *(اختياري)* — قدم شهادة TLS إذا كان مجمّعك يتطلبها.
+
+    <Frame>![تهيئة مجمّع OpenTelemetry](/images/crewai-otel-collector-opentelemetry.png)</Frame>
+  </Tab>
+  <Tab title="Datadog">
+    - **Datadog Site Domain** — مضيف OTLP لموقع Datadog الخاص بك فقط، دون بروتوكول أو مسار. يقوم CrewAI ببناء نقطة نهاية HTTPS OTLP الكاملة نيابةً عنك. استخدم المضيف المطابق لـ [موقع Datadog](https://docs.datadoghq.com/getting_started/site/) الخاص بك:
+      - `otlp.datadoghq.com` (US1)
+      - `otlp.us3.datadoghq.com` (US3)
+      - `otlp.us5.datadoghq.com` (US5)
+      - `otlp.datadoghq.eu` (EU1)
+      - `otlp.ap1.datadoghq.com` (AP1)
+    - **API Key** — مفتاح واجهة برمجة تطبيقات Datadog الخاص بك. راجع [كيفية إنشاء واحد](https://docs.datadoghq.com/account_management/api-app-keys/#api-keys).
+
+    يصدّر تكامل Datadog **التتبعات**.
+
+    <Frame>![تهيئة مجمّع Datadog](/images/crewai-otel-collector-datadog.png)</Frame>
+  </Tab>
+</Tabs>
+
+5. *(اختياري)* انقر على **Test Connection** للتحقق من قدرة CrewAI على الوصول إلى نقطة النهاية باستخدام بيانات الاعتماد التي قدمتها.
+6. انقر على **Save**.
 
 <Tip>
   يمكنك إضافة مجمّعات متعددة — على سبيل المثال، واحد للتتبعات وآخر للسجلات، أو الإرسال إلى واجهات خلفية مختلفة لأغراض مختلفة.

docs/ar/guides/advanced/customizing-prompts.mdx

@@ -161,6 +161,18 @@ crew = Crew(
 )
 ```
 
+<Note>
+يُحتفظ بـ `agent.i18n` للتوافق مع الإصدارات السابقة فقط، وقد تم إهماله. لتخصيص المطالبات أثناء التشغيل، مرّر `prompt_file` إلى `Crew`. وللوصول البرمجي المباشر إلى شرائح المطالبات، استخدم أداة i18n مباشرة:
+</Note>
+
+```python
+from crewai.utilities.i18n import get_i18n
+
+i18n = get_i18n("custom_prompts.json")
+format_slice = i18n.slice("format")
+tool_prompt = i18n.tools("ask_question")
+```
+
 #### الخيار 3: تعطيل مطالبات النظام لنماذج o1
 ```python
 agent = Agent(
@@ -208,6 +220,8 @@ agent = Agent(
 
 يدمج CrewAI بعد ذلك تخصيصاتك مع الإعدادات الافتراضية، فلا تحتاج لإعادة تعريف كل مطالبة. إليك الطريقة:
 
+بالنسبة للكود الذي يحتاج إلى قراءة شرائح المطالبات مباشرة، استخدم `crewai.utilities.i18n.get_i18n()` مع ملف المطالبات نفسه بدلًا من قراءة `agent.i18n`.
+
 ### مثال: تخصيص أساسي للمطالبات
 
 أنشئ ملف `custom_prompts.json` بالمطالبات التي تريد تعديلها. تأكد من إدراج جميع المطالبات عالية المستوى التي يجب أن يحتويها، وليس فقط تغييراتك:

docs/en/changelog.mdx

@@ -4,6 +4,106 @@ description: "Product updates, improvements, and bug fixes for CrewAI"
 icon: "clock"
 mode: "wide"
 ---
+<Update label="Jun 11, 2026">
+  ## v1.14.7
+
+  [View release on GitHub](https://github.com/crewAIInc/crewAI/releases/tag/1.14.7)
+
+  ## What's Changed
+
+  ### Features
+  - Add pluggable default backends for memory, knowledge, rag, and flow.
+  - Surface real finish_reason, sampling params, and response.id on LLM events.
+  - Type DSL triggers as route-aware decorators.
+  - Add chat API for conversational flows.
+  - Make locking backend overridable.
+  - Build FlowDefinition from Flow DSL metadata.
+  - Add native Snowflake Cortex LLM provider.
+  - Add crew trained agents file support.
+
+  ### Bug Fixes
+  - Fix checkpoint to rebuild custom BaseLLM as concrete LLM on restore.
+  - Gate restore on a flag to prevent live snapshots from replaying as resume.
+  - Scope runtime state per run to bound growth and isolate concurrent runs.
+  - Fix telemetry setup on crewai-login.
+  - Respect suppress_flow_events for method-execution events.
+  - Restore [project.scripts] in crewai package for uv tool install.
+  - Resolve pip-audit CVEs for aiohttp, docling, and docling-core.
+  - Fix file input not working reliably.
+  - Fix Snowflake Claude incomplete tool result histories.
+
+  ### Documentation
+  - Update changelog and version for v1.14.7.
+  - Update OpenTelemetry collector documentation.
+  - Update NVIDIA Nemotron LLM guide.
+  - Add Databricks integration guide.
+  - Add Snowflake integration guide.
+
+  ### Performance
+  - Improve crewai import speed by lazy-loading docling imports.
+
+  ### Refactoring
+  - Simplify flow condition evaluation to be stateless per event.
+  - Decouple convo logic from runtime and add a conversational_definition.
+  - Split `flow.py` into DSL, definition, and runtime.
+
+  ## Contributors
+
+  @Luzk, @alex-clawd, @devin-ai-integration[bot], @greysonlalonde, @gvieira, @jessemiller, @lorenzejay, @lucasgomide, @mattatcha, @vinibrsl
+
+</Update>
+
+<Update label="Jun 10, 2026">
+  ## v1.14.7rc2
+
+  [View release on GitHub](https://github.com/crewAIInc/crewAI/releases/tag/1.14.7rc2)
+
+  ## What's Changed
+
+  ### Bug Fixes
+  - Gate restore on a flag to prevent live snapshots from replaying as resume
+
+  ### Documentation
+  - Update changelog and version for v1.14.7rc1
+
+  ## Contributors
+
+  @greysonlalonde
+
+</Update>
+
+<Update label="Jun 10, 2026">
+  ## v1.14.7rc1
+
+  [View release on GitHub](https://github.com/crewAIInc/crewAI/releases/tag/1.14.7rc1)
+
+  ## What's Changed
+
+  ### Features
+  - Add `reset_runtime_state` to release accumulated bus state
+  - Handle supporting both custom prompts
+  - Decouple conversation logic from runtime and add a `conversational_definition`
+
+  ### Bug Fixes
+  - Fix scope of runtime state per run to bound growth and isolate concurrent runs
+  - Fix telemetry setup on `crewai-login`
+  - Fix respect for `suppress_flow_events` for method-execution events
+
+  ### Documentation
+  - Update OpenTelemetry images
+  - Update documentation to reflect new state of OpenTelemetry collector
+  - Update changelog and version for v1.14.7a4
+
+  ### Refactoring
+  - Simplify flow condition evaluation to be stateless per event
+  - Improve conversation routing cycle with one less route
+
+  ## Contributors
+
+  @greysonlalonde, @lorenzejay, @lucasgomide, @vinibrsl
+
+</Update>
+
 <Update label="Jun 09, 2026">
   ## v1.14.7a4
 

docs/en/concepts/flows.mdx

@@ -226,6 +226,49 @@ After the Flow has run, you can access the final state to see the updates made b
 By ensuring that the final method's output is returned and providing access to the state, CrewAI Flows make it easy to integrate the results of your AI workflows into larger applications or systems,
 while also maintaining and accessing the state throughout the Flow's execution.
 
+## Flow Usage Metrics
+
+After a Flow execution completes, you can access the `usage_metrics` property to view aggregated token usage across **every LLM call** made during the run — including calls from every Crew the Flow orchestrated, calls inside Agent tools, and bare `LLM.call(...)` invocations from Flow methods. This is the SDK-side equivalent of the totals shown in the CrewAI Enterprise UI.
+
+```python Code
+from crewai import LLM
+from crewai.flow.flow import Flow, listen, start
+
+class UsageMetricsFlow(Flow):
+    @start()
+    def run_first_crew(self):
+        self.state.first_result = FirstCrew().crew().kickoff()
+
+    @listen(run_first_crew)
+    def call_llm_directly(self):
+        # Bare LLM call — still counted by flow.usage_metrics
+        llm = LLM(model="openai/gpt-4o-mini")
+        self.state.summary = llm.call("Summarize the key takeaways.")
+
+    @listen(call_llm_directly)
+    def run_second_crew(self):
+        self.state.second_result = SecondCrew().crew().kickoff()
+
+flow = UsageMetricsFlow()
+flow.kickoff()
+
+print(flow.usage_metrics)
+# UsageMetrics(total_tokens=8579, prompt_tokens=6210, completion_tokens=2369,
+#              cached_prompt_tokens=0, reasoning_tokens=0,
+#              cache_creation_tokens=0, successful_requests=5)
+```
+
+<Note>
+  `flow.usage_metrics` is **not** the same as `flow.kickoff().token_usage`. The
+  latter returns the `CrewOutput.token_usage` of the **last** `@listen` method
+  that returned a `CrewOutput`, which means it only reflects the final Crew and
+  ignores prior Crews and bare `LLM.call(...)` invocations entirely. Use
+  `flow.usage_metrics` whenever you need the **full** token rollup for the Flow
+  execution.
+</Note>
+
+Each entry in the returned [`UsageMetrics`](https://github.com/crewAIInc/crewAI/blob/main/lib/crewai/src/crewai/types/usage_metrics.py) is the sum across all LLM calls made within a single `flow.kickoff()` invocation. Counters reset on the next `kickoff()` call (or on each iteration of `kickoff_for_each`), so successive runs don't double-count. The property is safe to read at any point after `kickoff()` completes; reading it during execution returns the partial total accumulated so far.
+
 ## Flow State Management
 
 Managing state effectively is crucial for building reliable and maintainable AI workflows. CrewAI Flows provides robust mechanisms for both unstructured and structured state management,

docs/en/enterprise/guides/capture_telemetry_logs.mdx

@@ -24,15 +24,39 @@ Telemetry data follows the [OpenTelemetry GenAI semantic conventions](https://op
 
 1. In CrewAI AMP, go to **Settings** > **OpenTelemetry Collectors**.
 2. Click **Add Collector**.
-3. Select an integration type — **OpenTelemetry Traces** or **OpenTelemetry Logs**.
-4. Configure the connection:
-   - **Endpoint** — Your collector's OTLP endpoint (e.g., `https://otel-collector.example.com:4317`).
-   - **Service Name** — A name to identify this service in your observability platform.
-   - **Custom Headers** *(optional)* — Add authentication or routing headers as key-value pairs.
-   - **Certificate** *(optional)* — Provide a TLS certificate if your collector requires one.
-5. Click **Save**.
+3. Select an integration:
+   - **OpenTelemetry Traces** and **OpenTelemetry Logs** — export to any OTLP-compatible collector or backend.
+   - **Datadog** — send traces straight to Datadog's OTLP intake, no separate collector or Datadog Agent required.
+4. Configure the connection. The fields depend on the integration you selected:
 
-<Frame>![OpenTelemetry Collector Configuration](/images/crewai-otel-collector-config.png)</Frame>
+<Tabs>
+  <Tab title="OpenTelemetry Traces / Logs">
+    **OpenTelemetry Traces** and **OpenTelemetry Logs** are separate integrations that share the same fields — pick the one matching the signal you want to export.
+
+    - **Endpoint** — Your collector's OTLP endpoint (e.g., `https://otel-collector.example.com:4317`).
+    - **Service Name** — A name to identify this service in your observability platform.
+    - **Custom Headers** *(optional)* — Add authentication or routing headers as key-value pairs.
+    - **Certificate** *(optional)* — Provide a TLS certificate if your collector requires one.
+
+    <Frame>![OpenTelemetry collector configuration](/images/crewai-otel-collector-opentelemetry.png)</Frame>
+  </Tab>
+  <Tab title="Datadog">
+    - **Datadog Site Domain** — Your Datadog site's OTLP host only, with no protocol or path. CrewAI builds the full HTTPS OTLP endpoint for you. Use the host that matches your [Datadog site](https://docs.datadoghq.com/getting_started/site/):
+      - `otlp.datadoghq.com` (US1)
+      - `otlp.us3.datadoghq.com` (US3)
+      - `otlp.us5.datadoghq.com` (US5)
+      - `otlp.datadoghq.eu` (EU1)
+      - `otlp.ap1.datadoghq.com` (AP1)
+    - **API Key** — Your Datadog API key. See [how to create one](https://docs.datadoghq.com/account_management/api-app-keys/#api-keys).
+
+    The Datadog integration exports **traces**.
+
+    <Frame>![Datadog collector configuration](/images/crewai-otel-collector-datadog.png)</Frame>
+  </Tab>
+</Tabs>
+
+5. *(optional)* Click **Test Connection** to verify CrewAI can reach the endpoint with the credentials you provided.
+6. Click **Save**.
 
 <Tip>
   You can add multiple collectors — for example, one for traces and another for logs, or send to different backends for different purposes.

docs/en/guides/advanced/customizing-prompts.mdx

@@ -161,6 +161,18 @@ crew = Crew(
 )
 ```
 
+<Note>
+`agent.i18n` is maintained only for backward compatibility and is deprecated. For runtime prompt customization, pass `prompt_file` to `Crew`. For programmatic access to prompt slices, use the i18n utility directly:
+</Note>
+
+```python
+from crewai.utilities.i18n import get_i18n
+
+i18n = get_i18n("custom_prompts.json")
+format_slice = i18n.slice("format")
+tool_prompt = i18n.tools("ask_question")
+```
+
 #### Option 3: Disable System Prompts for o1 Models
 ```python
 agent = Agent(
@@ -208,6 +220,8 @@ One straightforward approach is to create a JSON file for the prompts you want t
 
 CrewAI then merges your customizations with the defaults, so you don't have to redefine every prompt. Here's how:
 
+For code that needs to read prompt slices directly, use `crewai.utilities.i18n.get_i18n()` with the same prompt file instead of reading `agent.i18n`.
+
 ### Example: Basic Prompt Customization
 
 Create a `custom_prompts.json` file with the prompts you want to modify. Ensure you list all top-level prompts it should contain, not just your changes:

docs/ko/changelog.mdx

@@ -4,6 +4,106 @@ description: "CrewAI의 제품 업데이트, 개선 사항 및 버그 수정"
 icon: "clock"
 mode: "wide"
 ---
+<Update label="2026년 6월 11일">
+  ## v1.14.7
+
+  [GitHub 릴리스 보기](https://github.com/crewAIInc/crewAI/releases/tag/1.14.7)
+
+  ## 변경 사항
+
+  ### 기능
+  - 메모리, 지식, RAG 및 흐름에 대한 플러그 가능한 기본 백엔드를 추가했습니다.
+  - LLM 이벤트에서 실제 finish_reason, 샘플링 매개변수 및 response.id를 표시합니다.
+  - 경로 인식 장식자로서의 타입 DSL 트리거를 설정합니다.
+  - 대화 흐름을 위한 채팅 API를 추가했습니다.
+  - 잠금 백엔드를 재정의 가능하도록 만듭니다.
+  - Flow DSL 메타데이터에서 FlowDefinition을 빌드합니다.
+  - 네이티브 Snowflake Cortex LLM 공급자를 추가했습니다.
+  - 훈련된 에이전트 파일 지원을 추가했습니다.
+
+  ### 버그 수정
+  - 복원 시 사용자 정의 BaseLLM을 구체적인 LLM으로 재구성하도록 체크포인트를 수정했습니다.
+  - 라이브 스냅샷이 재개로 재생되지 않도록 플래그를 사용하여 복원을 제한합니다.
+  - 실행마다 런타임 상태의 범위를 설정하여 성장을 제한하고 동시 실행을 격리합니다.
+  - crewai-login에서 텔레메트리 설정을 수정했습니다.
+  - 메서드 실행 이벤트에 대해 suppress_flow_events를 존중합니다.
+  - uv 도구 설치를 위해 crewai 패키지에서 [project.scripts]를 복원합니다.
+  - aiohttp, docling 및 docling-core에 대한 pip-audit CVE를 해결합니다.
+  - 파일 입력이 신뢰할 수 없게 작동하는 문제를 수정했습니다.
+  - Snowflake Claude의 불완전한 도구 결과 기록을 수정했습니다.
+
+  ### 문서
+  - v1.14.7에 대한 변경 로그 및 버전을 업데이트했습니다.
+  - OpenTelemetry 수집기 문서를 업데이트했습니다.
+  - NVIDIA Nemotron LLM 가이드를 업데이트했습니다.
+  - Databricks 통합 가이드를 추가했습니다.
+  - Snowflake 통합 가이드를 추가했습니다.
+
+  ### 성능
+  - docling 가져오기를 지연 로딩하여 crewai 가져오기 속도를 개선했습니다.
+
+  ### 리팩토링
+  - 흐름 조건 평가를 이벤트별로 상태 비저장으로 단순화했습니다.
+  - 대화 논리를 런타임에서 분리하고 conversational_definition을 추가했습니다.
+  - `flow.py`를 DSL, 정의 및 런타임으로 분리했습니다.
+
+  ## 기여자
+
+  @Luzk, @alex-clawd, @devin-ai-integration[bot], @greysonlalonde, @gvieira, @jessemiller, @lorenzejay, @lucasgomide, @mattatcha, @vinibrsl
+
+</Update>
+
+<Update label="2026년 6월 10일">
+  ## v1.14.7rc2
+
+  [GitHub 릴리스 보기](https://github.com/crewAIInc/crewAI/releases/tag/1.14.7rc2)
+
+  ## 변경 사항
+
+  ### 버그 수정
+  - 라이브 스냅샷이 재개로 재생되는 것을 방지하기 위한 플래그에서 게이트 복원
+
+  ### 문서
+  - v1.14.7rc1에 대한 변경 로그 및 버전 업데이트
+
+  ## 기여자
+
+  @greysonlalonde
+
+</Update>
+
+<Update label="2026년 6월 10일">
+  ## v1.14.7rc1
+
+  [GitHub 릴리스 보기](https://github.com/crewAIInc/crewAI/releases/tag/1.14.7rc1)
+
+  ## 변경 사항
+
+  ### 기능
+  - 누적된 버스 상태를 해제하기 위해 `reset_runtime_state` 추가
+  - 사용자 정의 프롬프트를 모두 지원하도록 처리
+  - 대화 논리를 런타임과 분리하고 `conversational_definition` 추가
+
+  ### 버그 수정
+  - 실행당 런타임 상태의 범위를 수정하여 성장 제한 및 동시 실행 격리
+  - `crewai-login`에서 원격 측정 설정 수정
+  - 메서드 실행 이벤트에 대한 `suppress_flow_events` 존중 수정
+
+  ### 문서
+  - OpenTelemetry 이미지 업데이트
+  - OpenTelemetry 수집기의 새로운 상태를 반영하도록 문서 업데이트
+  - v1.14.7a4에 대한 변경 로그 및 버전 업데이트
+
+  ### 리팩토링
+  - 이벤트당 상태 비저장 방식으로 흐름 조건 평가 단순화
+  - 경로를 하나 줄여 대화 라우팅 사이클 개선
+
+  ## 기여자
+
+  @greysonlalonde, @lorenzejay, @lucasgomide, @vinibrsl
+
+</Update>
+
 <Update label="2026년 6월 9일">
   ## v1.14.7a4
 

docs/ko/concepts/flows.mdx

@@ -221,6 +221,48 @@ Flow가 실행된 후, 이러한 메소드들에 의해 수행된 업데이트
 최종 메소드의 출력이 반환되고 상태에 접근할 수 있도록 함으로써, CrewAI Flow는 AI 워크플로우의 결과를 더 큰 애플리케이션이나 시스템에 쉽게 통합할 수 있게 하며,
 Flow 실행 과정 전반에 걸쳐 상태를 유지하고 접근하면서도 이를 용이하게 만듭니다.
 
+## 플로우 사용 메트릭
+
+Flow 실행이 완료된 후, `usage_metrics` 속성에 접근하여 실행 동안 발생한 **모든 LLM 호출**의 토큰 사용량 집계를 확인할 수 있습니다. 여기에는 Flow가 오케스트레이션한 모든 Crew의 호출, Agent의 도구 내부에서 발생한 호출, 그리고 Flow 메서드에서 직접 호출한 `LLM.call(...)`이 모두 포함됩니다. 이는 CrewAI Enterprise UI에 표시되는 총량과 동등한 SDK 측 값입니다.
+
+```python Code
+from crewai import LLM
+from crewai.flow.flow import Flow, listen, start
+
+class UsageMetricsFlow(Flow):
+    @start()
+    def run_first_crew(self):
+        self.state.first_result = FirstCrew().crew().kickoff()
+
+    @listen(run_first_crew)
+    def call_llm_directly(self):
+        # 직접 LLM 호출 — flow.usage_metrics에서도 집계됩니다
+        llm = LLM(model="openai/gpt-4o-mini")
+        self.state.summary = llm.call("핵심 내용을 요약해 주세요.")
+
+    @listen(call_llm_directly)
+    def run_second_crew(self):
+        self.state.second_result = SecondCrew().crew().kickoff()
+
+flow = UsageMetricsFlow()
+flow.kickoff()
+
+print(flow.usage_metrics)
+# UsageMetrics(total_tokens=8579, prompt_tokens=6210, completion_tokens=2369,
+#              cached_prompt_tokens=0, reasoning_tokens=0,
+#              cache_creation_tokens=0, successful_requests=5)
+```
+
+<Note>
+  `flow.usage_metrics`는 `flow.kickoff().token_usage`와 **동일하지 않습니다**.
+  후자는 `CrewOutput`을 반환한 **마지막** `@listen` 메서드의
+  `CrewOutput.token_usage`만 반환하므로, 이전에 실행된 Crew들과 Flow 메서드에서
+  직접 호출한 `LLM.call(...)`은 전혀 포함되지 않습니다. Flow 실행에 대한
+  **전체** 토큰 집계가 필요할 때는 항상 `flow.usage_metrics`를 사용하십시오.
+</Note>
+
+반환되는 [`UsageMetrics`](https://github.com/crewAIInc/crewAI/blob/main/lib/crewai/src/crewai/types/usage_metrics.py)의 각 항목은 단일 `flow.kickoff()` 실행 동안 발생한 모든 LLM 호출의 합계입니다. 다음 `kickoff()` 호출(및 `kickoff_for_each`의 각 반복)에서 카운터가 초기화되므로 연속 실행이 이중으로 집계되지 않습니다. 이 속성은 `kickoff()` 완료 후 언제든지 안전하게 읽을 수 있으며, 실행 중에 읽으면 그 시점까지 누적된 부분 합계를 반환합니다.
+
 ## 플로우 상태 관리
 
 상태를 효과적으로 관리하는 것은 신뢰할 수 있고 유지 보수가 용이한 AI 워크플로를 구축하는 데 매우 중요합니다. CrewAI 플로우는 비정형 및 정형 상태 관리를 위한 강력한 메커니즘을 제공하여, 개발자가 자신의 애플리케이션에 가장 적합한 접근 방식을 선택할 수 있도록 합니다.

docs/ko/enterprise/guides/capture_telemetry_logs.mdx

@@ -24,15 +24,39 @@ CrewAI AMP는 배포에서 OpenTelemetry **트레이스**와 **로그**를 자
 
 1. CrewAI AMP에서 **Settings** > **OpenTelemetry Collectors**로 이동합니다.
 2. **Add Collector**를 클릭합니다.
-3. 통합 유형을 선택합니다 — **OpenTelemetry Traces** 또는 **OpenTelemetry Logs**.
-4. 연결을 구성합니다:
-   - **Endpoint** — 수집기의 OTLP 엔드포인트 (예: `https://otel-collector.example.com:4317`).
-   - **Service Name** — 관측 가능성 플랫폼에서 이 서비스를 식별하기 위한 이름.
-   - **Custom Headers** *(선택 사항)* — 인증 또는 라우팅 헤더를 키-값 쌍으로 추가합니다.
-   - **Certificate** *(선택 사항)* — 수집기에서 TLS 인증서가 필요한 경우 제공합니다.
-5. **Save**를 클릭합니다.
+3. 통합을 선택합니다:
+   - **OpenTelemetry Traces** 및 **OpenTelemetry Logs** — OTLP 호환 수집기 또는 백엔드로 내보냅니다.
+   - **Datadog** — 별도의 수집기나 Datadog Agent 없이 트레이스를 Datadog의 OTLP 인테이크로 직접 전송합니다.
+4. 연결을 구성합니다. 필드는 선택한 통합에 따라 달라집니다:
 
-<Frame>![OpenTelemetry 수집기 구성](/images/crewai-otel-collector-config.png)</Frame>
+<Tabs>
+  <Tab title="OpenTelemetry Traces / Logs">
+    **OpenTelemetry Traces**와 **OpenTelemetry Logs**는 동일한 필드를 공유하는 별개의 통합입니다 — 내보내려는 신호에 맞는 것을 선택하세요.
+
+    - **Endpoint** — 수집기의 OTLP 엔드포인트 (예: `https://otel-collector.example.com:4317`).
+    - **Service Name** — 관측 가능성 플랫폼에서 이 서비스를 식별하기 위한 이름.
+    - **Custom Headers** *(선택 사항)* — 인증 또는 라우팅 헤더를 키-값 쌍으로 추가합니다.
+    - **Certificate** *(선택 사항)* — 수집기에서 TLS 인증서가 필요한 경우 제공합니다.
+
+    <Frame>![OpenTelemetry 수집기 구성](/images/crewai-otel-collector-opentelemetry.png)</Frame>
+  </Tab>
+  <Tab title="Datadog">
+    - **Datadog Site Domain** — Datadog 사이트의 OTLP 호스트만 입력합니다 (프로토콜이나 경로 제외). CrewAI가 전체 HTTPS OTLP 엔드포인트를 자동으로 구성합니다. [Datadog 사이트](https://docs.datadoghq.com/getting_started/site/)에 맞는 호스트를 사용하세요:
+      - `otlp.datadoghq.com` (US1)
+      - `otlp.us3.datadoghq.com` (US3)
+      - `otlp.us5.datadoghq.com` (US5)
+      - `otlp.datadoghq.eu` (EU1)
+      - `otlp.ap1.datadoghq.com` (AP1)
+    - **API Key** — Datadog API 키입니다. [키 생성 방법](https://docs.datadoghq.com/account_management/api-app-keys/#api-keys)을 참고하세요.
+
+    Datadog 통합은 **트레이스**를 내보냅니다.
+
+    <Frame>![Datadog 수집기 구성](/images/crewai-otel-collector-datadog.png)</Frame>
+  </Tab>
+</Tabs>
+
+5. *(선택 사항)* **Test Connection**을 클릭하여 제공한 자격 증명으로 CrewAI가 엔드포인트에 연결할 수 있는지 확인합니다.
+6. **Save**를 클릭합니다.
 
 <Tip>
   여러 수집기를 추가할 수 있습니다 — 예를 들어, 트레이스용 하나와 로그용 하나를 추가하거나, 다른 목적을 위해 다른 백엔드로 전송할 수 있습니다.

docs/ko/guides/advanced/customizing-prompts.mdx

@@ -161,6 +161,18 @@ crew = Crew(
 )
 ```
 
+<Note>
+`agent.i18n`은 이전 버전과의 호환성을 위해서만 유지되며 사용이 중단될 예정입니다. 런타임 프롬프트 커스터마이징에는 `Crew`에 `prompt_file`을 전달하세요. 프롬프트 슬라이스를 코드에서 직접 읽어야 한다면 i18n 유틸리티를 직접 사용하세요:
+</Note>
+
+```python
+from crewai.utilities.i18n import get_i18n
+
+i18n = get_i18n("custom_prompts.json")
+format_slice = i18n.slice("format")
+tool_prompt = i18n.tools("ask_question")
+```
+
 #### 옵션 3: o1 모델에 대한 시스템 프롬프트 비활성화
 ```python
 agent = Agent(
@@ -208,6 +220,8 @@ agent = Agent(
 
 그러면 CrewAI가 기본값과 사용자가 지정한 내용을 병합하므로, 모든 프롬프트를 다시 정의할 필요가 없습니다. 방법은 다음과 같습니다:
 
+프롬프트 슬라이스를 코드에서 직접 읽어야 하는 경우에는 `agent.i18n`을 읽는 대신 동일한 프롬프트 파일로 `crewai.utilities.i18n.get_i18n()`을 사용하세요.
+
 ### 예시: 기본 프롬프트 커스터마이징
 
 수정하고 싶은 프롬프트를 포함하는 `custom_prompts.json` 파일을 생성하세요. 변경 사항만이 아니라 포함해야 하는 모든 최상위 프롬프트를 반드시 나열해야 합니다:
@@ -314,4 +328,4 @@ CrewAI에서의 저수준 prompt 커스터마이제이션은 매우 맞춤화되
 
 <Check>
 이제 CrewAI에서 고급 prompt 커스터마이징을 위한 기초를 갖추었습니다. 모델별 구조나 도메인별 제약에 맞춰 적용하든, 이러한 저수준 접근 방식은 agent 상호작용을 매우 전문적으로 조정할 수 있게 해줍니다.
-</Check>
+</Check>

docs/pt-BR/changelog.mdx

@@ -4,6 +4,106 @@ description: "Atualizações de produto, melhorias e correções do CrewAI"
 icon: "clock"
 mode: "wide"
 ---
+<Update label="11 jun 2026">
+  ## v1.14.7
+
+  [Ver release no GitHub](https://github.com/crewAIInc/crewAI/releases/tag/1.14.7)
+
+  ## O que Mudou
+
+  ### Recursos
+  - Adicionar backends padrão plugáveis para memória, conhecimento, rag e fluxo.
+  - Exibir o verdadeiro finish_reason, parâmetros de amostragem e response.id em eventos LLM.
+  - Tipar os gatilhos DSL como decoradores cientes de rotas.
+  - Adicionar API de chat para fluxos de conversa.
+  - Tornar o backend de bloqueio substituível.
+  - Construir FlowDefinition a partir de metadados Flow DSL.
+  - Adicionar provedor nativo Snowflake Cortex LLM.
+  - Adicionar suporte a arquivos de agentes treinados pela equipe.
+
+  ### Correções de Bugs
+  - Corrigir checkpoint para reconstruir BaseLLM personalizado como LLM concreto na restauração.
+  - Controlar a restauração com uma flag para evitar que snapshots ao vivo sejam reproduzidos como retomar.
+  - Escopar o estado de execução por execução para limitar o crescimento e isolar execuções concorrentes.
+  - Corrigir configuração de telemetria no crewai-login.
+  - Respeitar suppress_flow_events para eventos de execução de método.
+  - Restaurar [project.scripts] no pacote crewai para instalação da ferramenta uv.
+  - Resolver CVEs de pip-audit para aiohttp, docling e docling-core.
+  - Corrigir entrada de arquivo que não estava funcionando de forma confiável.
+  - Corrigir histórias de resultados de ferramentas incompletas do Snowflake Claude.
+
+  ### Documentação
+  - Atualizar changelog e versão para v1.14.7.
+  - Atualizar documentação do coletor OpenTelemetry.
+  - Atualizar guia do LLM NVIDIA Nemotron.
+  - Adicionar guia de integração do Databricks.
+  - Adicionar guia de integração do Snowflake.
+
+  ### Desempenho
+  - Melhorar a velocidade de importação do crewai através do carregamento preguiçoso de imports do docling.
+
+  ### Refatoração
+  - Simplificar a avaliação de condições de fluxo para ser sem estado por evento.
+  - Desacoplar a lógica de conversa da execução e adicionar uma conversational_definition.
+  - Dividir `flow.py` em DSL, definição e execução.
+
+  ## Contribuidores
+
+  @Luzk, @alex-clawd, @devin-ai-integration[bot], @greysonlalonde, @gvieira, @jessemiller, @lorenzejay, @lucasgomide, @mattatcha, @vinibrsl
+
+</Update>
+
+<Update label="10 jun 2026">
+  ## v1.14.7rc2
+
+  [Ver release no GitHub](https://github.com/crewAIInc/crewAI/releases/tag/1.14.7rc2)
+
+  ## O que Mudou
+
+  ### Correções de Bugs
+  - Restauração de portão em uma flag para evitar que snapshots ao vivo sejam reproduzidos como retomar
+
+  ### Documentação
+  - Atualizar changelog e versão para v1.14.7rc1
+
+  ## Contributors
+
+  @greysonlalonde
+
+</Update>
+
+<Update label="10 jun 2026">
+  ## v1.14.7rc1
+
+  [Ver release no GitHub](https://github.com/crewAIInc/crewAI/releases/tag/1.14.7rc1)
+
+  ## O que Mudou
+
+  ### Recursos
+  - Adicionar `reset_runtime_state` para liberar o estado acumulado do barramento
+  - Lidar com suporte a ambos os prompts personalizados
+  - Desacoplar a lógica de conversa do tempo de execução e adicionar uma `conversational_definition`
+
+  ### Correções de Bugs
+  - Corrigir o escopo do estado de tempo de execução por execução para limitar o crescimento e isolar execuções concorrentes
+  - Corrigir a configuração de telemetria em `crewai-login`
+  - Corrigir o respeito a `suppress_flow_events` para eventos de execução de método
+
+  ### Documentação
+  - Atualizar imagens do OpenTelemetry
+  - Atualizar a documentação para refletir o novo estado do coletor OpenTelemetry
+  - Atualizar o changelog e a versão para v1.14.7a4
+
+  ### Refatoração
+  - Simplificar a avaliação da condição de fluxo para ser sem estado por evento
+  - Melhorar o ciclo de roteamento de conversas com uma rota a menos
+
+  ## Contribuidores
+
+  @greysonlalonde, @lorenzejay, @lucasgomide, @vinibrsl
+
+</Update>
+
 <Update label="09 jun 2026">
   ## v1.14.7a4
 

docs/pt-BR/concepts/flows.mdx

@@ -219,6 +219,49 @@ Após o término da execução, é possível acessar o estado final e observar a
 Ao garantir que a saída do método final seja retornada e oferecer acesso ao estado, o CrewAI Flows facilita a integração dos resultados dos seus workflows de IA em aplicações maiores,
 além de permitir o gerenciamento e o acesso ao estado durante toda a execução do Flow.
 
+## Métricas de Uso do Flow
+
+Após a execução de um Flow, você pode acessar a propriedade `usage_metrics` para visualizar o consumo agregado de tokens em **todas as chamadas de LLM** realizadas durante a execução — incluindo chamadas das Crews orquestradas pelo Flow, chamadas dentro de tools de Agents, e invocações diretas de `LLM.call(...)` feitas a partir de métodos do Flow. Esse é o equivalente, do lado do SDK, ao total exibido na interface do CrewAI Enterprise.
+
+```python Code
+from crewai import LLM
+from crewai.flow.flow import Flow, listen, start
+
+class UsageMetricsFlow(Flow):
+    @start()
+    def run_first_crew(self):
+        self.state.first_result = FirstCrew().crew().kickoff()
+
+    @listen(run_first_crew)
+    def call_llm_directly(self):
+        # Chamada direta de LLM — também contabilizada por flow.usage_metrics
+        llm = LLM(model="openai/gpt-4o-mini")
+        self.state.summary = llm.call("Resuma os principais pontos.")
+
+    @listen(call_llm_directly)
+    def run_second_crew(self):
+        self.state.second_result = SecondCrew().crew().kickoff()
+
+flow = UsageMetricsFlow()
+flow.kickoff()
+
+print(flow.usage_metrics)
+# UsageMetrics(total_tokens=8579, prompt_tokens=6210, completion_tokens=2369,
+#              cached_prompt_tokens=0, reasoning_tokens=0,
+#              cache_creation_tokens=0, successful_requests=5)
+```
+
+<Note>
+  `flow.usage_metrics` **não** é o mesmo que `flow.kickoff().token_usage`. Este
+  último retorna apenas o `CrewOutput.token_usage` do **último** método
+  `@listen` que retornou um `CrewOutput`, ou seja, reflete somente a Crew
+  final e ignora completamente as Crews anteriores e quaisquer chamadas
+  diretas de `LLM.call(...)`. Use `flow.usage_metrics` sempre que precisar do
+  rollup **completo** de tokens da execução do Flow.
+</Note>
+
+Cada campo do [`UsageMetrics`](https://github.com/crewAIInc/crewAI/blob/main/lib/crewai/src/crewai/types/usage_metrics.py) retornado representa a soma de todas as chamadas de LLM feitas em uma única invocação de `flow.kickoff()`. Os contadores são resetados a cada novo `kickoff()` (e em cada iteração de `kickoff_for_each`), de modo que execuções sucessivas não duplicam o total. A propriedade é segura para ser lida em qualquer momento após o `kickoff()`; lê-la durante a execução retorna o total parcial acumulado até aquele instante.
+
 ## Gerenciamento de Estado em Flows
 
 Gerenciar o estado de forma eficaz é fundamental para construir fluxos de trabalho de IA confiáveis e de fácil manutenção. O CrewAI Flows oferece mecanismos robustos para o gerenciamento de estado tanto não estruturado quanto estruturado,

docs/pt-BR/enterprise/guides/capture_telemetry_logs.mdx

@@ -24,15 +24,39 @@ Os dados de telemetria seguem as [convenções semânticas GenAI do OpenTelemetr
 
 1. No CrewAI AMP, vá para **Settings** > **OpenTelemetry Collectors**.
 2. Clique em **Add Collector**.
-3. Selecione um tipo de integração — **OpenTelemetry Traces** ou **OpenTelemetry Logs**.
-4. Configure a conexão:
-   - **Endpoint** — O endpoint OTLP do seu coletor (por exemplo, `https://otel-collector.example.com:4317`).
-   - **Service Name** — Um nome para identificar este serviço na sua plataforma de observabilidade.
-   - **Custom Headers** *(opcional)* — Adicione headers de autenticação ou roteamento como pares chave-valor.
-   - **Certificate** *(opcional)* — Forneça um certificado TLS se o seu coletor exigir um.
-5. Clique em **Save**.
+3. Selecione uma integração:
+   - **OpenTelemetry Traces** e **OpenTelemetry Logs** — exporte para qualquer coletor ou backend compatível com OTLP.
+   - **Datadog** — envie traces diretamente para a ingestão OTLP do Datadog, sem precisar de um coletor separado ou do Datadog Agent.
+4. Configure a conexão. Os campos dependem da integração selecionada:
 
-<Frame>![Configuração do Coletor OpenTelemetry](/images/crewai-otel-collector-config.png)</Frame>
+<Tabs>
+  <Tab title="OpenTelemetry Traces / Logs">
+    **OpenTelemetry Traces** e **OpenTelemetry Logs** são integrações separadas que compartilham os mesmos campos — escolha a que corresponde ao sinal que você quer exportar.
+
+    - **Endpoint** — O endpoint OTLP do seu coletor (por exemplo, `https://otel-collector.example.com:4317`).
+    - **Service Name** — Um nome para identificar este serviço na sua plataforma de observabilidade.
+    - **Custom Headers** *(opcional)* — Adicione headers de autenticação ou roteamento como pares chave-valor.
+    - **Certificate** *(opcional)* — Forneça um certificado TLS se o seu coletor exigir um.
+
+    <Frame>![Configuração do coletor OpenTelemetry](/images/crewai-otel-collector-opentelemetry.png)</Frame>
+  </Tab>
+  <Tab title="Datadog">
+    - **Datadog Site Domain** — Apenas o host OTLP do seu site Datadog, sem protocolo ou caminho. O CrewAI monta o endpoint HTTPS OTLP completo para você. Use o host correspondente ao seu [site Datadog](https://docs.datadoghq.com/getting_started/site/):
+      - `otlp.datadoghq.com` (US1)
+      - `otlp.us3.datadoghq.com` (US3)
+      - `otlp.us5.datadoghq.com` (US5)
+      - `otlp.datadoghq.eu` (EU1)
+      - `otlp.ap1.datadoghq.com` (AP1)
+    - **API Key** — Sua chave de API do Datadog. Veja [como criar uma](https://docs.datadoghq.com/account_management/api-app-keys/#api-keys).
+
+    A integração com o Datadog exporta **traces**.
+
+    <Frame>![Configuração do coletor Datadog](/images/crewai-otel-collector-datadog.png)</Frame>
+  </Tab>
+</Tabs>
+
+5. *(opcional)* Clique em **Test Connection** para verificar se o CrewAI consegue acessar o endpoint com as credenciais fornecidas.
+6. Clique em **Save**.
 
 <Tip>
   Você pode adicionar múltiplos coletores — por exemplo, um para traces e outro para logs, ou enviar para diferentes backends para diferentes propósitos.

docs/pt-BR/guides/advanced/customizing-prompts.mdx

@@ -161,6 +161,18 @@ crew = Crew(
 )
 ```
 
+<Note>
+`agent.i18n` é mantido apenas para compatibilidade retroativa e está obsoleto. Para customização de prompts em tempo de execução, passe `prompt_file` para `Crew`. Para acesso programático aos slices de prompt, use diretamente o utilitário de i18n:
+</Note>
+
+```python
+from crewai.utilities.i18n import get_i18n
+
+i18n = get_i18n("custom_prompts.json")
+format_slice = i18n.slice("format")
+tool_prompt = i18n.tools("ask_question")
+```
+
 #### Opção 3: Desativar Prompts de Sistema para Modelos o1
 ```python
 agent = Agent(
@@ -208,6 +220,8 @@ Uma abordagem direta é criar um arquivo JSON para os prompts que deseja sobresc
 
 O CrewAI então mescla suas customizações com os padrões, assim você não precisa redefinir todos os prompts. Veja como:
 
+Para código que precisa ler slices de prompt diretamente, use `crewai.utilities.i18n.get_i18n()` com o mesmo arquivo de prompts em vez de ler `agent.i18n`.
+
 ### Exemplo: Customização Básica de Prompt
 
 Crie um arquivo `custom_prompts.json` com os prompts que deseja modificar. Certifique-se de listar todos os prompts de nível superior que ele deve conter, não apenas suas alterações:

lib/cli/pyproject.toml

@@ -8,7 +8,7 @@ authors = [
 ]
 requires-python = ">=3.10, <3.14"
 dependencies = [
-    "crewai-core==1.14.7a4",
+    "crewai-core==1.14.7",
     "click>=8.1.7,<9",
     "pydantic>=2.11.9,<2.13",
     "pydantic-settings~=2.10.1",

lib/cli/src/crewai_cli/__init__.py

@@ -1 +1 @@
-__version__ = "1.14.7a4"
+__version__ = "1.14.7"

lib/cli/src/crewai_cli/templates/crew/pyproject.toml

@@ -5,7 +5,7 @@ description = "{{name}} using crewAI"
 authors = [{ name = "Your Name", email = "you@example.com" }]
 requires-python = ">=3.10,<3.14"
 dependencies = [
-    "crewai[tools]==1.14.7a4"
+    "crewai[tools]==1.14.7"
 ]
 
 [project.scripts]

lib/cli/src/crewai_cli/templates/flow/pyproject.toml

@@ -5,7 +5,7 @@ description = "{{name}} using crewAI"
 authors = [{ name = "Your Name", email = "you@example.com" }]
 requires-python = ">=3.10,<3.14"
 dependencies = [
-    "crewai[tools]==1.14.7a4"
+    "crewai[tools]==1.14.7"
 ]
 
 [project.scripts]

lib/cli/src/crewai_cli/templates/tool/pyproject.toml

@@ -5,7 +5,7 @@ description = "Power up your crews with {{folder_name}}"
 readme = "README.md"
 requires-python = ">=3.10,<3.14"
 dependencies = [
-    "crewai[tools]==1.14.7a4"
+    "crewai[tools]==1.14.7"
 ]
 
 [tool.crewai]

lib/crewai-core/src/crewai_core/__init__.py

@@ -1 +1 @@
-__version__ = "1.14.7a4"
+__version__ = "1.14.7"

lib/crewai-core/src/crewai_core/telemetry.py

@@ -17,7 +17,7 @@ import contextlib
 import logging
 import os
 import threading
-from typing import Any, Final
+from typing import Any, ClassVar, Final
 
 from opentelemetry import trace
 from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
@@ -27,7 +27,7 @@ from opentelemetry.sdk.trace.export import (
     BatchSpanProcessor,
     SpanExportResult,
 )
-from opentelemetry.trace import Span, Status, StatusCode
+from opentelemetry.trace import ProxyTracerProvider, Span, Status, StatusCode
 from typing_extensions import Self
 
 
@@ -72,8 +72,8 @@ class Telemetry:
     and event-bus signal handlers (see ``crewai.telemetry.telemetry``).
     """
 
-    _instance = None
-    _lock = threading.Lock()
+    _instance: ClassVar[Self | None] = None
+    _lock: ClassVar[threading.Lock] = threading.Lock()
 
     def __new__(cls) -> Self:
         if cls._instance is None:
@@ -149,6 +149,10 @@ class Telemetry:
         if self.ready and not self.trace_set:
             try:
                 with suppress_warnings():
+                    existing_provider = trace.get_tracer_provider()
+                    if not isinstance(existing_provider, ProxyTracerProvider):
+                        self.trace_set = True
+                        return
                     trace.set_tracer_provider(self.provider)
                     self.trace_set = True
             except Exception as e:

lib/crewai-core/tests/test_smoke.py

@@ -14,6 +14,7 @@ from crewai_core import (
     version,
 )
 import pytest
+from opentelemetry.sdk.trace import TracerProvider
 
 
 def test_version_returns_string() -> None:
@@ -94,3 +95,36 @@ def test_user_data_decline_blocks(
 def test_unused_var_warning_silenced() -> None:
     # Touch os to keep the import (used by env-var fixtures above)
     assert os.environ is not None
+
+
+def test_core_telemetry_skips_duplicate_tracer_provider(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    from crewai_core.telemetry import Telemetry
+
+    Telemetry._instance = None
+    monkeypatch.delenv("OTEL_SDK_DISABLED", raising=False)
+    monkeypatch.delenv("CREWAI_DISABLE_TELEMETRY", raising=False)
+    monkeypatch.delenv("CREWAI_DISABLE_TRACKING", raising=False)
+
+    monkeypatch.setattr(
+        "crewai_core.telemetry.trace.get_tracer_provider",
+        lambda: TracerProvider(),
+    )
+
+    called = False
+
+    def fail_if_called(provider: object) -> None:
+        nonlocal called
+        called = True
+
+    monkeypatch.setattr(
+        "crewai_core.telemetry.trace.set_tracer_provider",
+        fail_if_called,
+    )
+
+    telemetry = Telemetry()
+    telemetry.set_tracer()
+
+    assert called is False
+    assert telemetry.trace_set is True

lib/crewai-files/src/crewai_files/__init__.py

@@ -152,4 +152,4 @@ __all__ = [
     "wrap_file_source",
 ]
 
-__version__ = "1.14.7a4"
+__version__ = "1.14.7"

lib/crewai-tools/pyproject.toml

@@ -10,7 +10,7 @@ requires-python = ">=3.10, <3.14"
 dependencies = [
     "pytube~=15.0.0",
     "requests>=2.33.0,<3",
-    "crewai==1.14.7a4",
+    "crewai==1.14.7",
     "tiktoken>=0.8.0,<0.13",
     "beautifulsoup4~=4.13.4",
     "python-docx~=1.2.0",

lib/crewai-tools/src/crewai_tools/__init__.py

@@ -330,4 +330,4 @@ __all__ = [
     "ZapierActionTools",
 ]
 
-__version__ = "1.14.7a4"
+__version__ = "1.14.7"

lib/crewai/pyproject.toml

@@ -8,8 +8,8 @@ authors = [
 ]
 requires-python = ">=3.10, <3.14"
 dependencies = [
-    "crewai-core==1.14.7a4",
-    "crewai-cli==1.14.7a4",
+    "crewai-core==1.14.7",
+    "crewai-cli==1.14.7",
     # Core Dependencies
     "pydantic>=2.11.9,<2.13",
     "openai>=2.30.0,<3",
@@ -54,7 +54,7 @@ Repository = "https://github.com/crewAIInc/crewAI"
 
 [project.optional-dependencies]
 tools = [
-    "crewai-tools==1.14.7a4",
+    "crewai-tools==1.14.7",
 ]
 embeddings = [
     "tiktoken>=0.8.0,<0.13"

lib/crewai/src/crewai/__init__.py

@@ -48,7 +48,7 @@ def _suppress_pydantic_deprecation_warnings() -> None:
 
 _suppress_pydantic_deprecation_warnings()
 
-__version__ = "1.14.7a4"
+__version__ = "1.14.7"
 
 _LAZY_IMPORTS: dict[str, tuple[str, str]] = {
     "Memory": ("crewai.memory.unified_memory", "Memory"),

lib/crewai/src/crewai/agents/agent_builder/base_agent.py

@@ -46,6 +46,7 @@ from crewai.state.checkpoint_config import CheckpointConfig, _coerce_checkpoint
 from crewai.tools.base_tool import BaseTool, Tool
 from crewai.types.callback import SerializableCallable
 from crewai.utilities.config import process_config
+from crewai.utilities.i18n import I18N, get_i18n
 from crewai.utilities.logger import Logger
 from crewai.utilities.rpm_controller import RPMController
 from crewai.utilities.string_utils import interpolate_only
@@ -81,6 +82,7 @@ _LLM_TYPE_REGISTRY: dict[str, str] = {
 def _validate_llm_ref(value: Any) -> Any:
     if isinstance(value, dict):
         import importlib
+        import inspect
 
         llm_type = value.get("llm_type")
         if not llm_type or llm_type not in _LLM_TYPE_REGISTRY:
@@ -91,6 +93,12 @@ def _validate_llm_ref(value: Any) -> Any:
         dotted = _LLM_TYPE_REGISTRY[llm_type]
         mod_path, cls_name = dotted.rsplit(".", 1)
         cls = getattr(importlib.import_module(mod_path), cls_name)
+        if inspect.isabstract(cls):
+            from crewai.llm import LLM
+
+            return LLM(
+                **{k: v for k, v in value.items() if v is not None and k != "llm_type"}
+            )
         return cls(**value)
     return value
 
@@ -186,6 +194,7 @@ class BaseAgent(BaseModel, ABC, metaclass=AgentMeta):
         tools (list[Any] | None): Tools at the agent's disposal.
         max_iter (int): Maximum iterations for an agent to execute a task.
         agent_executor: An instance of the CrewAgentExecutor class.
+        i18n (I18N): Internationalization settings.
         llm (Any): Language model that will run the agent.
         crew (Any): Crew to which the agent belongs.
 
@@ -265,6 +274,14 @@ class BaseAgent(BaseModel, ABC, metaclass=AgentMeta):
             _serialize_executor_ref, return_type=dict | None, when_used="json"
         ),
     ] = Field(default=None, description="An instance of the CrewAgentExecutor class.")
+    i18n: I18N = Field(
+        default_factory=get_i18n,
+        description="Internationalization settings.",
+        deprecated=(
+            "Agent.i18n is deprecated and will be removed in a future release. "
+            "Use crewai.utilities.i18n.get_i18n() or Crew(prompt_file=...) instead."
+        ),
+    )
 
     llm: Annotated[
         str | BaseLLM | None,

lib/crewai/src/crewai/context.py

@@ -117,8 +117,10 @@ def capture_execution_context(
     )
 
 
-def apply_execution_context(ctx: ExecutionContext) -> None:
+def apply_execution_context(ctx: ExecutionContext | dict[str, Any]) -> None:
     """Write an ExecutionContext back into the ContextVars."""
+    if isinstance(ctx, dict):
+        ctx = ExecutionContext.model_validate(ctx)
     _current_task_id.set(ctx.current_task_id)
     current_flow_request_id.set(ctx.flow_request_id)
     current_flow_id.set(ctx.flow_id)

lib/crewai/src/crewai/crew.py

@@ -1013,6 +1013,7 @@ class Crew(FlowTrackable, BaseModel):
         )
         token = attach(baggage_ctx)
 
+        runtime_scope = crewai_event_bus._enter_runtime_scope()
         try:
             inputs = prepare_kickoff(self, inputs, input_files)
 
@@ -1048,6 +1049,7 @@ class Crew(FlowTrackable, BaseModel):
                 self._memory.drain_writes()
             clear_files(self.id)
             detach(token)
+            crewai_event_bus._exit_runtime_scope(runtime_scope)
 
     def _post_kickoff(self, result: CrewOutput) -> CrewOutput:
         return result
@@ -1223,6 +1225,7 @@ class Crew(FlowTrackable, BaseModel):
         )
         token = attach(baggage_ctx)
 
+        runtime_scope = crewai_event_bus._enter_runtime_scope()
         try:
             inputs = prepare_kickoff(self, inputs, input_files)
 
@@ -1256,6 +1259,7 @@ class Crew(FlowTrackable, BaseModel):
         finally:
             clear_files(self.id)
             detach(token)
+            crewai_event_bus._exit_runtime_scope(runtime_scope)
 
     async def akickoff_for_each(
         self,

lib/crewai/src/crewai/events/event_bus.py

@@ -80,6 +80,17 @@ def is_replaying() -> bool:
     return _replaying.get()
 
 
+_runtime_state_var: contextvars.ContextVar[RuntimeState | None] = (
+    contextvars.ContextVar("crewai_runtime_state", default=None)
+)
+_registered_entity_ids_var: contextvars.ContextVar[set[int] | None] = (
+    contextvars.ContextVar("crewai_registered_entity_ids", default=None)
+)
+_runtime_scope_depth: contextvars.ContextVar[int] = contextvars.ContextVar(
+    "crewai_runtime_scope_depth", default=0
+)
+
+
 class CrewAIEventsBus:
     """Singleton event bus for handling events in CrewAI.
 
@@ -116,7 +127,6 @@ class CrewAIEventsBus:
     _futures_lock: threading.Lock
     _executor_initialized: bool
     _has_pending_events: bool
-    _runtime_state: RuntimeState | None
 
     def __new__(cls) -> Self:
         """Create or return the singleton instance.
@@ -151,8 +161,6 @@ class CrewAIEventsBus:
         self._console = ConsoleFormatter()
         self._executor_initialized = False
         self._has_pending_events = False
-        self._runtime_state: RuntimeState | None = None
-        self._registered_entity_ids: set[int] = set()
 
     def _ensure_executor_initialized(self) -> None:
         """Lazily initialize the thread pool executor and event loop.
@@ -281,6 +289,51 @@ class CrewAIEventsBus:
         """The RuntimeState currently attached to the bus, if any."""
         return self._runtime_state
 
+    @property
+    def _runtime_state(self) -> RuntimeState | None:
+        return _runtime_state_var.get()
+
+    @_runtime_state.setter
+    def _runtime_state(self, value: RuntimeState | None) -> None:
+        _runtime_state_var.set(value)
+
+    @property
+    def _registered_entity_ids(self) -> set[int]:
+        ids = _registered_entity_ids_var.get()
+        if ids is None:
+            ids = set()
+            _registered_entity_ids_var.set(ids)
+        return ids
+
+    @_registered_entity_ids.setter
+    def _registered_entity_ids(self, value: set[int]) -> None:
+        _registered_entity_ids_var.set(value)
+
+    def reset_runtime_state(self) -> None:
+        """Detach the RuntimeState and clear the entity registry."""
+        self._runtime_state = None
+        self._registered_entity_ids = set()
+
+    def _enter_runtime_scope(self) -> bool:
+        depth = _runtime_scope_depth.get()
+        _runtime_scope_depth.set(depth + 1)
+        if depth != 0:
+            return False
+        if _runtime_state_var.get() is None:
+            from crewai import RuntimeState
+
+            if RuntimeState is not None:
+                _runtime_state_var.set(RuntimeState(root=[]))
+            _registered_entity_ids_var.set(set())
+        return True
+
+    def _exit_runtime_scope(self, outermost: bool) -> None:
+        depth = _runtime_scope_depth.get()
+        _runtime_scope_depth.set(depth - 1 if depth > 0 else 0)
+        if outermost:
+            _runtime_state_var.set(None)
+            _registered_entity_ids_var.set(None)
+
     def register_entity(self, entity: Any) -> None:
         """Add an entity to the RuntimeState, creating it if needed.
 
@@ -349,6 +402,7 @@ class CrewAIEventsBus:
         source: Any,
         event: BaseEvent,
         handlers: SyncHandlerSet,
+        state: RuntimeState | None,
     ) -> None:
         """Call provided synchronous handlers.
 
@@ -356,8 +410,8 @@ class CrewAIEventsBus:
             source: The emitting object
             event: The event instance
             handlers: Frozenset of sync handlers to call
+            state: The RuntimeState captured on the emitting context
         """
-        state = self._runtime_state
         errors: list[tuple[SyncHandler, Exception]] = [
             (handler, error)
             for handler in handlers
@@ -376,6 +430,7 @@ class CrewAIEventsBus:
         source: Any,
         event: BaseEvent,
         handlers: AsyncHandlerSet,
+        state: RuntimeState | None,
     ) -> None:
         """Asynchronously call provided async handlers.
 
@@ -383,8 +438,8 @@ class CrewAIEventsBus:
             source: The object that emitted the event
             event: The event instance
             handlers: Frozenset of async handlers to call
+            state: The RuntimeState captured on the emitting context
         """
-        state = self._runtime_state
 
         async def _call(handler: AsyncHandler) -> Any:
             if _get_param_count(handler) >= 3:
@@ -399,7 +454,9 @@ class CrewAIEventsBus:
                     f"[CrewAIEventsBus] Async handler error in {getattr(handler, '__name__', handler)}: {result}"
                 )
 
-    async def _emit_with_dependencies(self, source: Any, event: BaseEvent) -> None:
+    async def _emit_with_dependencies(
+        self, source: Any, event: BaseEvent, state: RuntimeState | None
+    ) -> None:
         """Emit an event with dependency-aware handler execution.
 
         Handlers are grouped into execution levels based on their dependencies.
@@ -450,18 +507,18 @@ class CrewAIEventsBus:
 
             if level_sync:
                 if event_type is LLMStreamChunkEvent:
-                    self._call_handlers(source, event, level_sync)
+                    self._call_handlers(source, event, level_sync, state)
                 else:
                     ctx = contextvars.copy_context()
                     future = self._sync_executor.submit(
-                        ctx.run, self._call_handlers, source, event, level_sync
+                        ctx.run, self._call_handlers, source, event, level_sync, state
                     )
                     await asyncio.get_running_loop().run_in_executor(
                         None, future.result
                     )
 
             if level_async:
-                await self._acall_handlers(source, event, level_async)
+                await self._acall_handlers(source, event, level_async, state)
 
     def _register_source(self, source: Any) -> None:
         """Register the source entity in RuntimeState if applicable."""
@@ -556,21 +613,23 @@ class CrewAIEventsBus:
         self._ensure_executor_initialized()
         self._has_pending_events = True
 
+        state = self._runtime_state
+
         if has_dependencies:
             return self._track_future(
                 asyncio.run_coroutine_threadsafe(
-                    self._emit_with_dependencies(source, event),
+                    self._emit_with_dependencies(source, event, state),
                     self._loop,
                 )
             )
 
         if sync_handlers:
             if event_type is LLMStreamChunkEvent:
-                self._call_handlers(source, event, sync_handlers)
+                self._call_handlers(source, event, sync_handlers, state)
             else:
                 ctx = contextvars.copy_context()
                 sync_future = self._sync_executor.submit(
-                    ctx.run, self._call_handlers, source, event, sync_handlers
+                    ctx.run, self._call_handlers, source, event, sync_handlers, state
                 )
                 if not async_handlers:
                     return self._track_future(sync_future)
@@ -578,7 +637,7 @@ class CrewAIEventsBus:
         if async_handlers:
             return self._track_future(
                 asyncio.run_coroutine_threadsafe(
-                    self._acall_handlers(source, event, async_handlers),
+                    self._acall_handlers(source, event, async_handlers, state),
                     self._loop,
                 )
             )
@@ -590,21 +649,22 @@ class CrewAIEventsBus:
         source: Any,
         event: BaseEvent,
         handlers: AsyncHandlerSet,
+        state: RuntimeState | None,
     ) -> None:
         """Call async handlers with the replaying flag set on the loop thread."""
         token = _replaying.set(True)
         try:
-            await self._acall_handlers(source, event, handlers)
+            await self._acall_handlers(source, event, handlers, state)
         finally:
             _replaying.reset(token)
 
     async def _emit_with_dependencies_replaying(
-        self, source: Any, event: BaseEvent
+        self, source: Any, event: BaseEvent, state: RuntimeState | None
     ) -> None:
         """Dependency-aware dispatch with the replaying flag set."""
         token = _replaying.set(True)
         try:
-            await self._emit_with_dependencies(source, event)
+            await self._emit_with_dependencies(source, event, state)
         finally:
             _replaying.reset(token)
 
@@ -638,12 +698,13 @@ class CrewAIEventsBus:
         self._ensure_executor_initialized()
         self._has_pending_events = True
 
+        state = self._runtime_state
         token = _replaying.set(True)
         try:
             if has_dependencies:
                 return self._track_future(
                     asyncio.run_coroutine_threadsafe(
-                        self._emit_with_dependencies_replaying(source, event),
+                        self._emit_with_dependencies_replaying(source, event, state),
                         self._loop,
                     )
                 )
@@ -651,7 +712,7 @@ class CrewAIEventsBus:
             if sync_handlers:
                 ctx = contextvars.copy_context()
                 sync_future = self._sync_executor.submit(
-                    ctx.run, self._call_handlers, source, event, sync_handlers
+                    ctx.run, self._call_handlers, source, event, sync_handlers, state
                 )
                 self._track_future(sync_future)
                 if not async_handlers:
@@ -659,7 +720,9 @@ class CrewAIEventsBus:
 
             return self._track_future(
                 asyncio.run_coroutine_threadsafe(
-                    self._acall_handlers_replaying(source, event, async_handlers),
+                    self._acall_handlers_replaying(
+                        source, event, async_handlers, state
+                    ),
                     self._loop,
                 )
             )
@@ -727,7 +790,9 @@ class CrewAIEventsBus:
             async_handlers = self._async_handlers.get(event_type, frozenset())
 
         if async_handlers:
-            await self._acall_handlers(source, event, async_handlers)
+            await self._acall_handlers(
+                source, event, async_handlers, self._runtime_state
+            )
 
     def register_handler(
         self,

lib/crewai/src/crewai/events/listeners/tracing/trace_listener.py

@@ -292,7 +292,7 @@ class TraceCollectionListener(BaseEventListener):
         @event_bus.on(CrewKickoffCompletedEvent)
         def on_crew_completed(source: Any, event: CrewKickoffCompletedEvent) -> None:
             self._handle_trace_event("crew_kickoff_completed", source, event)
-            if self.batch_manager.defer_session_finalization:
+            if self._should_defer_session_finalization():
                 return
             if self._nested_in_flow_execution():
                 return
@@ -306,7 +306,7 @@ class TraceCollectionListener(BaseEventListener):
         @event_bus.on(CrewKickoffFailedEvent)
         def on_crew_failed(source: Any, event: CrewKickoffFailedEvent) -> None:
             self._handle_trace_event("crew_kickoff_failed", source, event)
-            if self.batch_manager.defer_session_finalization:
+            if self._should_defer_session_finalization():
                 return
             if self._nested_in_flow_execution():
                 return
@@ -734,7 +734,7 @@ class TraceCollectionListener(BaseEventListener):
             if not self.batch_manager.is_batch_initialized():
                 return
             # Multi-turn flows defer batch finalization to finalize_session_traces().
-            if self.batch_manager.defer_session_finalization:
+            if self._should_defer_session_finalization():
                 return
             self.batch_manager.finalize_batch()
 
@@ -745,6 +745,15 @@ class TraceCollectionListener(BaseEventListener):
 
         return current_flow_id.get() is not None
 
+    def _should_defer_session_finalization(self) -> bool:
+        """True when the active trace belongs to a deferred flow session."""
+        from crewai.flow.flow_context import current_flow_defer_trace_finalization
+
+        return (
+            self.batch_manager.defer_session_finalization
+            or current_flow_defer_trace_finalization.get()
+        )
+
     def _flow_owns_trace_batch(self) -> bool:
         """True when an in-flight conversational flow already owns the trace batch."""
         if self.batch_manager.batch_owner_type == "flow":
@@ -780,12 +789,17 @@ class TraceCollectionListener(BaseEventListener):
     def _try_initialize_flow_batch_from_context(self, event: Any) -> bool:
         """Claim a flow trace batch when an action event fires inside kickoff.
 
-        When ``suppress_flow_events=True``, console panels are hidden but
-        ``FlowStartedEvent`` and method lifecycle events still emit; if no
-        batch exists yet, LLM/tool events must not fall back to implicit crew
-        batches.
+        When ``suppress_flow_events=True`` (infrastructure flows such as
+        ``AgentExecutor`` and the memory flows), flow and method lifecycle
+        events are not emitted, so the batch is claimed from the flow context
+        (``current_flow_id``) to keep LLM/tool events from falling back to an
+        implicit crew batch.
         """
-        from crewai.flow.flow_context import current_flow_id, current_flow_name
+        from crewai.flow.flow_context import (
+            current_flow_defer_trace_finalization,
+            current_flow_id,
+            current_flow_name,
+        )
 
         flow_id = current_flow_id.get()
         if flow_id is None:
@@ -801,6 +815,8 @@ class TraceCollectionListener(BaseEventListener):
         }
         self.batch_manager.batch_owner_type = "flow"
         self.batch_manager.batch_owner_id = flow_id
+        if current_flow_defer_trace_finalization.get():
+            self.batch_manager.defer_session_finalization = True
         self._initialize_batch(user_context, execution_metadata)
         return True
 

lib/crewai/src/crewai/events/types/flow_events.py

@@ -1,6 +1,6 @@
 from typing import Any, Literal
 
-from pydantic import BaseModel, ConfigDict
+from pydantic import BaseModel, ConfigDict, field_serializer
 
 from crewai.events.base_events import BaseEvent
 
@@ -57,6 +57,10 @@ class MethodExecutionFailedEvent(FlowEvent):
 
     model_config = ConfigDict(arbitrary_types_allowed=True)
 
+    @field_serializer("error")
+    def _serialize_error(self, error: Exception) -> str:
+        return str(error)
+
 
 class MethodExecutionPausedEvent(FlowEvent):
     """Event emitted when a flow method is paused waiting for human feedback.

lib/crewai/src/crewai/experimental/conversational_mixin.py

@@ -1,15 +1,17 @@
-"""Conversational graph + helpers as a mixin for ``Flow`` (experimental).
+"""Conversational graph + helpers as an experimental Flow extension.
 
-The experimental conversational chat surface lives here as a mixin so that
-``crewai.flow.runtime`` stays focused on the execution engine. ``Flow``
-inherits from ``_ConversationalMixin``; the methods only register on
-subclasses that opt in via ``conversational = True`` (enforced by the
-``_conversational_only`` marker + ``FlowMeta`` gating in
-``crewai.flow.runtime``).
+The conversational chat surface remains experimental and may change before the
+v2 graduation path. It lives here so ``crewai.flow.runtime`` can stay focused
+on the execution engine. ``crewai.flow.flow`` composes this mixin onto the
+public ``Flow`` class for backwards compatibility.
+
+The built-in conversational graph only registers for subclasses that opt in
+with ``conversational = True``. Static conversational metadata is projected
+into ``FlowDefinition.conversational`` via the Python DSL builder.
 
 Import surface:
-    - :class:`_ConversationalMixin` — internal; ``Flow`` mixes it in. Users
-      don't import it directly.
+    - :class:`_ConversationalMixin` — internal; the public ``Flow`` class
+      composes it in. Users don't import it directly.
     - The data types this mixin uses live in
       :mod:`crewai.experimental.conversational`.
 """
@@ -20,7 +22,7 @@ from collections.abc import Callable, Mapping, Sequence
 from enum import Enum
 import json
 import logging
-from typing import TYPE_CHECKING, Any, ClassVar, Literal, cast
+from typing import TYPE_CHECKING, Any, ClassVar, Literal, TypeVar, cast
 
 from pydantic import BaseModel, Field, create_model
 
@@ -44,26 +46,69 @@ from crewai.flow.conversation import (
     get_conversation_messages,
     receive_user_message as _receive_user_message,
 )
-from crewai.flow.dsl import listen, router, start
+from crewai.flow.dsl import listen, start
+from crewai.flow.dsl._utils import _set_flow_method_definition
+from crewai.flow.flow_definition import FlowMethodDefinition
 from crewai.utilities.types import LLMMessage
 
 
 if TYPE_CHECKING:
-    from crewai.flow.runtime import Flow
     from crewai.llms.base_llm import BaseLLM
 
 
 logger = logging.getLogger(__name__)
 
 
-class _ConversationalMixin:
-    """Built-in conversational graph for ``Flow`` (gated on ``conversational``).
+def _iter_condition_labels(condition: Any) -> set[str]:
+    if isinstance(condition, str):
+        return {condition}
+    if isinstance(condition, dict):
+        labels: set[str] = set()
+        for value in condition.values():
+            if isinstance(value, list):
+                for item in value:
+                    labels.update(_iter_condition_labels(item))
+            else:
+                labels.update(_iter_condition_labels(value))
+        return labels
+    return set()
 
-    Mixed into ``Flow`` so its execution engine (``runtime.py``) stays focused
-    on running graphs. The methods here only register on subclasses that set
-    ``conversational = True``; non-chat flows see them as inert attributes.
+
+def _conversation_start_router(func: Callable[..., Any]) -> Any:
+    wrapper = start()(func)
+    _set_flow_method_definition(
+        cast(Any, wrapper),
+        FlowMethodDefinition(start=True, router=True),
+    )
+    return wrapper
+
+
+class _ConversationalMixin:
+    """Experimental conversational graph for ``Flow``.
+
+    This mixin owns chat behavior and runtime hooks. Non-chat flows see these
+    methods as inert attributes unless they opt in with ``conversational = True``.
     """
 
+    # === EXPERIMENTAL: conversational mode ===
+    # When ``conversational = True`` on a Flow subclass, this mixin's built-in
+    # graph registers and ``handle_turn`` / ``chat`` become chat entry points.
+    conversational: ClassVar[bool] = False
+    conversational_config: ClassVar[ConversationConfig | None] = None
+    builtin_routes: ClassVar[tuple[str, ...]] = ("converse", "end")
+    internal_routes: ClassVar[tuple[str, ...]] = ("answer_from_history",)
+    builtin_route_descriptions: ClassVar[dict[str, str]] = {
+        "converse": (
+            "Ordinary chat, follow-ups, summaries, clarifications, and "
+            "questions answerable from prior conversation history."
+        ),
+        "end": ("User signals the conversation is finished (goodbye, exit, done)."),
+        "answer_from_history": (
+            "Answer directly from prior conversation history without invoking "
+            "tools, agents, or custom routes."
+        ),
+    }
+
     # The metaclass + state attributes referenced below live on ``Flow`` —
     # this mixin is never instantiated standalone. These type-only
     # declarations exist so static analyzers don't flag attribute access.
@@ -71,22 +116,15 @@ class _ConversationalMixin:
     # (otherwise mypy flags "Cannot override instance variable with class
     # variable" when Flow declares them as ``ClassVar``).
     if TYPE_CHECKING:
-        conversational: ClassVar[bool]
-        conversational_config: ClassVar[ConversationConfig | None]
-        builtin_routes: ClassVar[tuple[str, ...]]
-        internal_routes: ClassVar[tuple[str, ...]]
-        builtin_route_descriptions: ClassVar[dict[str, str]]
-        # Registry ClassVars populated by ``FlowMeta`` at class creation.
-        _listeners: ClassVar[dict[Any, Any]]
-
         # Instance attrs from ``Flow``.
         state: Any
         name: str | None
         _completed_methods: set[Any]
         _method_outputs: list[Any]
-        _pending_and_listeners: dict[Any, Any]
+        _pending_events: dict[Any, Any]
         _method_call_counts: dict[Any, int]
         _is_execution_resuming: bool
+        _conversation_messages: list[LLMMessage]
         _pending_user_message: str | dict[str, Any] | None
         _pending_intents: Sequence[str] | None
         _pending_intent_llm: str | BaseLLM | None
@@ -97,8 +135,8 @@ class _ConversationalMixin:
         def _collapse_to_outcome(
             self,
             feedback: str,
-            outcomes: tuple[str, ...],
-            llm: str | BaseLLM | Any,
+            outcomes: Sequence[str],
+            llm: str | BaseLLM,
         ) -> str:
             pass
 
@@ -108,23 +146,24 @@ class _ConversationalMixin:
         def kickoff(self, *args: Any, **kwargs: Any) -> Any:
             pass
 
-    @start()
-    @_conversational_only
     def conversation_start(self) -> str | None:
-        """Internal Flow entrypoint that hands the user message to the router.
+        """Return the current user message for conversational route selection.
 
-        In conversational mode, ``Flow.kickoff_async`` runs all ``@start``
-        methods sequentially and this one is registered last, so any user
-        ``@start`` methods (e.g. permission loading) have already finished
-        before the returned value triggers ``route_conversation``.
+        This remains as a plain overridable helper for compatibility. It is not
+        registered as a Flow method; ``route_conversation`` is the synthetic
+        built-in start/router that begins a conversational turn.
         """
         state = cast(ConversationState, self.state)
         return state.current_user_message
 
-    @router(conversation_start)
+    @_conversation_start_router
     @_conversational_only
     def route_conversation(self) -> str:
         """Route the current turn to a listener label."""
+        if "conversation_start" not in {
+            str(method_name) for method_name in self._completed_methods
+        }:
+            self.conversation_start()
         state = cast(ConversationState, self.state)
         context = self.build_router_context()
         previous_intent = state.last_intent
@@ -238,8 +277,8 @@ class _ConversationalMixin:
         state = cast(ConversationState, self.state)
         sid = session_id or state.id
 
-        # Stash the pending turn so ``_apply_pending_conversational_turn``
-        # picks it up AFTER persist restore.
+        # Stash the pending turn so the kickoff extension hook picks it up
+        # after persist restore.
         self._pending_user_message = message
         self._pending_intents = list(intents) if intents else None
         self._pending_intent_llm = intent_llm
@@ -286,7 +325,7 @@ class _ConversationalMixin:
         callers can customize prompts or exercise the loop without patching
         builtins.
         """
-        if not getattr(type(self), "conversational", False):
+        if not self._is_conversational_enabled():
             raise ValueError("Flow.chat() is only available on conversational flows")
 
         exit_set = {command.lower() for command in exit_commands}
@@ -491,14 +530,14 @@ class _ConversationalMixin:
         **extra: Any,
     ) -> None:
         """Append a message to conversation history (legacy ChatState path)."""
-        _append_conversation_message(cast("Flow[Any]", self), role, content, **extra)
+        _append_conversation_message(cast(Any, self), role, content, **extra)
 
     @property
     def conversation_messages(self) -> list[LLMMessage]:
         """Message history from state, coerced to LLM-shaped dicts."""
         return [
             message_to_llm_dict(message)
-            for message in get_conversation_messages(cast("Flow[Any]", self))
+            for message in get_conversation_messages(cast(Any, self))
         ]
 
     def receive_user_message(
@@ -514,7 +553,7 @@ class _ConversationalMixin:
         ``state.messages`` and preserve ``last_intent`` across turns.
         Non-conversational flows fall through to the legacy helper.
         """
-        if self.conversational:
+        if self._is_conversational_enabled():
             state = cast(ConversationState, self.state)
             state.messages.append(ConversationMessage(role="user", content=text))
             self._emit_conversation_message_added(
@@ -535,9 +574,7 @@ class _ConversationalMixin:
                 return intent
             return text
 
-        return _receive_user_message(
-            cast("Flow[Any]", self), text, outcomes=outcomes, llm=llm
-        )
+        return _receive_user_message(cast(Any, self), text, outcomes=outcomes, llm=llm)
 
     def classify_intent(
         self,
@@ -561,27 +598,104 @@ class _ConversationalMixin:
     def _conversation_config(self) -> ConversationConfig | None:
         return getattr(type(self), "conversational_config", None)
 
+    @property
+    def _conversation_definition(self) -> Any | None:
+        return self._conversation_flow_definition().conversational
+
+    def _conversation_flow_definition(self) -> Any:
+        flow_definition = getattr(type(self), "flow_definition", None)
+        if not callable(flow_definition):
+            raise AttributeError(
+                f"{type(self).__name__} does not expose flow_definition()"
+            )
+        return flow_definition()
+
+    @classmethod
+    def _conversational_definition(cls) -> Any | None:
+        flow_definition = getattr(cls, "flow_definition", None)
+        if not callable(flow_definition):
+            return None
+        return flow_definition().conversational
+
+    @classmethod
+    def _is_conversational(cls) -> bool:
+        definition = cls._conversational_definition()
+        return bool(definition and definition.enabled)
+
+    def _is_conversational_enabled(self) -> bool:
+        definition = self._conversation_definition
+        return bool(definition and definition.enabled)
+
+    def _initialize_runtime_extension_attrs(self) -> None:
+        if not isinstance(getattr(self, "_conversation_messages", None), list):
+            object.__setattr__(self, "_conversation_messages", [])
+        if not hasattr(self, "_pending_user_message"):
+            object.__setattr__(self, "_pending_user_message", None)
+        if not hasattr(self, "_pending_intents"):
+            object.__setattr__(self, "_pending_intents", None)
+        if not hasattr(self, "_pending_intent_llm"):
+            object.__setattr__(self, "_pending_intent_llm", None)
+
+    def _create_default_extension_state(self) -> ConversationState | None:
+        initial_state_t = getattr(self, "_initial_state_t", None)
+        if type(self)._is_conversational() and (
+            not hasattr(self, "_initial_state_t")
+            or isinstance(initial_state_t, TypeVar)
+        ):
+            return ConversationState()
+        return None
+
+    def _should_apply_pending_kickoff_context(self) -> bool:
+        return (
+            type(self)._is_conversational() and self._pending_user_message is not None
+        )
+
+    def _apply_pending_kickoff_context(self) -> None:
+        self._apply_pending_conversational_turn()
+
+    def _order_start_methods_for_kickoff(
+        self,
+        start_methods: list[Any],
+    ) -> tuple[list[Any], bool]:
+        if not type(self)._is_conversational():
+            return start_methods, False
+
+        route_conversation = "route_conversation"
+        if route_conversation not in {str(method) for method in start_methods}:
+            return start_methods, False
+
+        ordered_starts = [
+            method for method in start_methods if str(method) != route_conversation
+        ]
+        ordered_starts.append(
+            next(
+                method for method in start_methods if str(method) == route_conversation
+            )
+        )
+        return ordered_starts, True
+
     def _should_defer_trace_finalization(self) -> bool:
         """Whether per-turn ``FlowFinished`` + ``finalize_batch`` should be skipped.
 
         True when either:
           - ``flow.defer_trace_finalization`` is set on the instance, OR
-          - the class-level ``ConversationConfig.defer_trace_finalization``
-            on a conversational subclass is True.
+          - the static conversational definition enables deferred finalization.
 
         Either source enables the deferred-session pattern. The caller
         eventually invokes ``finalize_session_traces()`` to close the batch.
         """
         if getattr(self, "defer_trace_finalization", False):
             return True
-        config = self._conversation_config
-        return bool(config and config.defer_trace_finalization)
+        definition = self._conversation_definition
+        return bool(
+            definition and definition.enabled and definition.defer_trace_finalization
+        )
 
     def _reset_turn_execution_state(self) -> None:
         """Clear per-execution tracking so the next turn re-runs the graph."""
         self._completed_methods.clear()
         self._method_outputs.clear()
-        self._pending_and_listeners.clear()
+        self._pending_events.clear()
         self._method_call_counts.clear()
         self._clear_or_listeners()
         self._is_execution_resuming = False
@@ -733,11 +847,12 @@ class _ConversationalMixin:
         router_config: RouterConfig | None,
     ) -> dict[str, str]:
         label_to_method: dict[str, str] = {}
-        for listener_name, condition in self._listeners.items():
-            if isinstance(condition, tuple):
-                _, trigger_labels = condition
-                for trigger_label in trigger_labels:
-                    label_to_method.setdefault(str(trigger_label), str(listener_name))
+        flow_definition = self._conversation_flow_definition()
+        for listener_name, method_definition in flow_definition.methods.items():
+            if method_definition.listen is None or method_definition.router:
+                continue
+            for trigger_label in _iter_condition_labels(method_definition.listen):
+                label_to_method.setdefault(trigger_label, listener_name)
 
         routes = self._effective_routes(router_config)
         overrides = (
@@ -788,21 +903,31 @@ class _ConversationalMixin:
 
     def _valid_route_labels(self) -> set[str]:
         labels: set[str] = set()
-        for condition in self._listeners.values():
-            if isinstance(condition, tuple):
-                _, methods = condition
-                labels.update(str(method) for method in methods)
+        flow_definition = self._conversation_flow_definition()
+        for method_definition in flow_definition.methods.values():
+            if method_definition.listen is None or method_definition.router:
+                continue
+            labels.update(_iter_condition_labels(method_definition.listen))
         return labels
 
     def _effective_routes(self, router_config: RouterConfig | None = None) -> set[str]:
         custom_routes = set(router_config.routes or ()) if router_config else set()
+        definition = self._conversation_definition
+        builtin_routes = (
+            tuple(definition.builtin_routes)
+            if definition is not None
+            else self.builtin_routes
+        )
+        internal_routes = (
+            tuple(definition.internal_routes)
+            if definition is not None
+            else self.internal_routes
+        )
         if not custom_routes:
             custom_routes = (
-                self._valid_route_labels()
-                - set(self.builtin_routes)
-                - set(self.internal_routes)
+                self._valid_route_labels() - set(builtin_routes) - set(internal_routes)
             )
-        return custom_routes | set(self.builtin_routes)
+        return custom_routes | set(builtin_routes)
 
     def _default_conversation_llm(self) -> Any | None:
         config = self._conversation_config
@@ -931,12 +1056,15 @@ class _ConversationalMixin:
 
         trace_listener = TraceCollectionListener()
         batch_manager = trace_listener.batch_manager
-        if batch_manager.batch_owner_type == "flow":
-            if trace_listener.first_time_handler.is_first_time:
-                trace_listener.first_time_handler.mark_events_collected()
-                trace_listener.first_time_handler.handle_execution_completion()
-            else:
-                batch_manager.finalize_batch()
+        try:
+            if batch_manager.batch_owner_type == "flow":
+                if trace_listener.first_time_handler.is_first_time:
+                    trace_listener.first_time_handler.mark_events_collected()
+                    trace_listener.first_time_handler.handle_execution_completion()
+                else:
+                    batch_manager.finalize_batch()
+        finally:
+            batch_manager.defer_session_finalization = False
 
 
 __all__ = ["_ConversationalMixin"]

lib/crewai/src/crewai/flow/conversational_definition.py

@@ -0,0 +1,48 @@
+"""Static conversational Flow definition models.
+
+This module is part of the serializable Flow Definition contract. It should
+only contain static data shapes. Experimental conversational runtime behavior
+continues to live in ``crewai.experimental.conversational_mixin``.
+"""
+
+from __future__ import annotations
+
+from typing import Any, Literal
+
+from pydantic import BaseModel, Field
+
+
+class FlowConversationalRouterDefinition(BaseModel):
+    """Static conversational router configuration."""
+
+    prompt: str | None = None
+    response_format: Any = None
+    llm: Any = None
+    routes: list[str] | None = None
+    route_descriptions: dict[str, str] | None = None
+    default_intent: str | None = "converse"
+    fallback_intent: str | None = "converse"
+    intent_field: str = "intent"
+
+
+class FlowConversationalDefinition(BaseModel):
+    """Static conversational Flow configuration."""
+
+    enabled: bool = False
+    system_prompt: str | None = None
+    llm: Any = None
+    router: FlowConversationalRouterDefinition | None = None
+    answer_from_history_prompt: str | None = None
+    default_intents: list[str] | None = None
+    intent_llm: Any = None
+    answer_from_history_llm: Any = None
+    visible_agent_outputs: list[str] | Literal["all"] | None = None
+    defer_trace_finalization: bool = True
+    builtin_routes: list[str] = Field(default_factory=lambda: ["converse", "end"])
+    internal_routes: list[str] = Field(default_factory=lambda: ["answer_from_history"])
+
+
+__all__ = [
+    "FlowConversationalDefinition",
+    "FlowConversationalRouterDefinition",
+]

lib/crewai/src/crewai/flow/dsl/_utils.py

@@ -9,6 +9,8 @@ from typing_extensions import TypeIs
 
 from crewai.flow.flow_definition import (
     FlowConfigDefinition,
+    FlowConversationalDefinition,
+    FlowConversationalRouterDefinition,
     FlowDefinition,
     FlowDefinitionDiagnostic,
     FlowHumanFeedbackDefinition,
@@ -27,6 +29,13 @@ R = TypeVar("R")
 logger = logging.getLogger(__name__)
 
 _FLOW_METHOD_DEFINITION_ATTR = "__flow_method_definition__"
+_FLOW_METHOD_METADATA_ATTRS = [
+    "__conversational_only__",
+    "__flow_method_definition__",
+    "__flow_persistence_config__",
+    "__human_feedback_config__",
+    "_human_feedback_llm",
+]
 
 
 def is_flow_method(obj: Any) -> TypeIs[FlowMethod[Any, Any]]:
@@ -42,6 +51,39 @@ def _should_include_flow_method(flow_class: type, method: Any) -> bool:
     return True
 
 
+def _is_conversational_flow(flow_class: type) -> bool:
+    return bool(getattr(flow_class, "conversational", False))
+
+
+def _get_inherited_conversational_method(
+    flow_class: type,
+    attr_name: str,
+) -> Any | None:
+    if not _is_conversational_flow(flow_class):
+        return None
+
+    for base in flow_class.__mro__[1:]:
+        inherited = base.__dict__.get(attr_name)
+        if inherited is None:
+            continue
+        if getattr(inherited, "__conversational_only__", False) and is_flow_method(
+            inherited
+        ):
+            return inherited
+    return None
+
+
+def _stamp_inherited_conversational_metadata(
+    method: Any,
+    inherited: Any,
+) -> Any:
+    for attr in _FLOW_METHOD_METADATA_ATTRS:
+        if hasattr(inherited, attr):
+            setattr(method, attr, getattr(inherited, attr))
+    method.__is_flow_method__ = True
+    return method
+
+
 def _set_flow_method_definition(
     wrapper: FlowMethod[P, R],
     definition: FlowMethodDefinition,
@@ -135,6 +177,8 @@ def _build_state_definition(
     from pydantic import BaseModel as PydanticBaseModel
 
     state_value = getattr(flow_class, "_initial_state_t", None)
+    if isinstance(state_value, TypeVar):
+        state_value = None
     initial_state = getattr(flow_class, "initial_state", None)
     if initial_state is not None:
         state_value = initial_state
@@ -230,6 +274,98 @@ def _build_persistence_definition(
     )
 
 
+def _build_conversational_router_definition(
+    router_config: Any,
+    diagnostics: list[FlowDefinitionDiagnostic],
+    path: str,
+) -> FlowConversationalRouterDefinition | None:
+    if router_config is None:
+        return None
+
+    routes = getattr(router_config, "routes", None)
+    return FlowConversationalRouterDefinition(
+        prompt=getattr(router_config, "prompt", None),
+        response_format=_serialize_static_value(
+            getattr(router_config, "response_format", None),
+            diagnostics,
+            f"{path}.response_format",
+        ),
+        llm=_serialize_static_value(
+            getattr(router_config, "llm", None), diagnostics, f"{path}.llm"
+        ),
+        routes=[str(route) for route in routes] if routes is not None else None,
+        route_descriptions=getattr(router_config, "route_descriptions", None),
+        default_intent=getattr(router_config, "default_intent", "converse"),
+        fallback_intent=getattr(router_config, "fallback_intent", "converse"),
+        intent_field=str(getattr(router_config, "intent_field", "intent")),
+    )
+
+
+def _build_conversational_definition(
+    flow_class: type,
+    diagnostics: list[FlowDefinitionDiagnostic],
+) -> FlowConversationalDefinition | None:
+    if not _is_conversational_flow(flow_class):
+        return None
+
+    config = getattr(flow_class, "conversational_config", None)
+    builtin_routes = getattr(flow_class, "builtin_routes", ("converse", "end"))
+    internal_routes = getattr(
+        flow_class,
+        "internal_routes",
+        ("answer_from_history",),
+    )
+    if config is None:
+        return FlowConversationalDefinition(
+            enabled=True,
+            builtin_routes=[str(route) for route in builtin_routes],
+            internal_routes=[str(route) for route in internal_routes],
+        )
+
+    default_intents = getattr(config, "default_intents", None)
+    visible_agent_outputs = getattr(config, "visible_agent_outputs", None)
+    return FlowConversationalDefinition(
+        enabled=True,
+        system_prompt=getattr(config, "system_prompt", None),
+        llm=_serialize_static_value(
+            getattr(config, "llm", None), diagnostics, "conversational.llm"
+        ),
+        router=_build_conversational_router_definition(
+            getattr(config, "router", None),
+            diagnostics,
+            "conversational.router",
+        ),
+        answer_from_history_prompt=getattr(config, "answer_from_history_prompt", None),
+        default_intents=(
+            [str(intent) for intent in default_intents]
+            if default_intents is not None
+            else None
+        ),
+        intent_llm=_serialize_static_value(
+            getattr(config, "intent_llm", None),
+            diagnostics,
+            "conversational.intent_llm",
+        ),
+        answer_from_history_llm=_serialize_static_value(
+            getattr(config, "answer_from_history_llm", None),
+            diagnostics,
+            "conversational.answer_from_history_llm",
+        ),
+        visible_agent_outputs=(
+            "all"
+            if visible_agent_outputs == "all"
+            else [str(output) for output in visible_agent_outputs]
+            if visible_agent_outputs is not None
+            else None
+        ),
+        defer_trace_finalization=bool(
+            getattr(config, "defer_trace_finalization", True)
+        ),
+        builtin_routes=[str(route) for route in builtin_routes],
+        internal_routes=[str(route) for route in internal_routes],
+    )
+
+
 def _build_method_definition(
     method: Any,
     diagnostics: list[FlowDefinitionDiagnostic],
@@ -270,6 +406,29 @@ def _iter_flow_methods(flow_class: type) -> dict[str, Any]:
             flow_class, attr_value
         ):
             methods[attr_name] = attr_value
+            continue
+
+        inherited = _get_inherited_conversational_method(flow_class, attr_name)
+        if inherited is not None and callable(attr_value):
+            methods[attr_name] = _stamp_inherited_conversational_metadata(
+                attr_value, inherited
+            )
+
+    if _is_conversational_flow(flow_class):
+        for base in reversed(flow_class.__mro__[1:]):
+            for attr_name, raw_value in base.__dict__.items():
+                if attr_name.startswith("_") or attr_name in methods:
+                    continue
+                if not getattr(raw_value, "__conversational_only__", False):
+                    continue
+                try:
+                    attr_value = getattr(flow_class, attr_name)
+                except AttributeError:
+                    continue
+                if is_flow_method(attr_value) and _should_include_flow_method(
+                    flow_class, attr_value
+                ):
+                    methods[attr_name] = attr_value
 
     # A wrapped method whose name collides with a base Flow model field
     # (e.g. ``checkpoint``) is absorbed by Pydantic as a field; the underlying
@@ -314,6 +473,7 @@ def _build_flow_definition_from_class(
         state=_build_state_definition(flow_class, diagnostics),
         config=_build_config_definition(flow_class, diagnostics),
         persist=_build_persistence_definition(flow_class, diagnostics, "persist"),
+        conversational=_build_conversational_definition(flow_class, diagnostics),
         methods=methods,
         diagnostics=diagnostics,
     )

lib/crewai/src/crewai/flow/flow.py

@@ -6,15 +6,22 @@ The implementation now lives in three modules, split by concern:
   ``@router``, ``or_`` / ``and_``) and Python Flow class projection
 - ``crewai.flow.flow_definition`` -- the serializable Flow Definition contract
 - ``crewai.flow.runtime`` -- the Flow execution engine and state
+- ``crewai.experimental.conversational_mixin`` -- experimental conversational
+  runtime extension composed onto the public ``Flow`` class
 
 Prefer importing from those modules in new code; this module preserves the
 historical ``crewai.flow.flow`` import path.
 """
 
+from typing import Any, TypeVar
+
+from pydantic import BaseModel
+
+from crewai.experimental.conversational_mixin import _ConversationalMixin
 from crewai.flow.dsl import and_, listen, or_, router, start
 from crewai.flow.runtime import (
     _INITIAL_STATE_CLASS_MARKER,
-    Flow,
+    Flow as RuntimeFlow,
     FlowMeta,
     FlowState,
     LockedDictProxy,
@@ -23,6 +30,13 @@ from crewai.flow.runtime import (
 )
 
 
+T = TypeVar("T", bound=dict[str, Any] | BaseModel)
+
+
+class Flow(_ConversationalMixin, RuntimeFlow[T]):
+    """Public Flow class with experimental conversational extension behavior."""
+
+
 __all__ = [
     "_INITIAL_STATE_CLASS_MARKER",
     "Flow",

lib/crewai/src/crewai/flow/flow_context.py

@@ -15,6 +15,10 @@ current_flow_id: contextvars.ContextVar[str | None] = contextvars.ContextVar(
     "flow_id", default=None
 )
 
+current_flow_defer_trace_finalization: contextvars.ContextVar[bool] = (
+    contextvars.ContextVar("flow_defer_trace_finalization", default=False)
+)
+
 current_flow_method_name: contextvars.ContextVar[str] = contextvars.ContextVar(
     "flow_method_name", default="unknown"
 )

lib/crewai/src/crewai/flow/flow_definition.py

@@ -16,6 +16,11 @@ from typing import Any, Literal as TypingLiteral
 from pydantic import BaseModel, ConfigDict, Field
 import yaml
 
+from crewai.flow.conversational_definition import (
+    FlowConversationalDefinition,
+    FlowConversationalRouterDefinition,
+)
+
 
 logger = logging.getLogger(__name__)
 
@@ -23,6 +28,8 @@ FlowDefinitionCondition = str | dict[str, Any]
 
 __all__ = [
     "FlowConfigDefinition",
+    "FlowConversationalDefinition",
+    "FlowConversationalRouterDefinition",
     "FlowDefinition",
     "FlowDefinitionCondition",
     "FlowDefinitionDiagnostic",
@@ -115,6 +122,7 @@ class FlowDefinition(BaseModel):
     state: FlowStateDefinition | None = None
     config: FlowConfigDefinition = Field(default_factory=FlowConfigDefinition)
     persist: FlowPersistenceDefinition | None = None
+    conversational: FlowConversationalDefinition | None = None
     methods: dict[str, FlowMethodDefinition] = Field(default_factory=dict)
     diagnostics: list[FlowDefinitionDiagnostic] = Field(default_factory=list)
 

lib/crewai/src/crewai/flow/types.py

@@ -16,7 +16,7 @@ R = TypeVar("R", covariant=True)
 FlowMethodName = NewType("FlowMethodName", str)
 PendingListenerKey = NewType(
     "PendingListenerKey",
-    Annotated[str, "nested flow conditions use 'listener_name:object_id'"],
+    Annotated[str, "listener method name, or 'start:<method>' for conditional starts"],
 )
 
 

lib/crewai/src/crewai/memory/recall_flow.py

@@ -259,8 +259,9 @@ class RecallFlow(Flow[RecallState]):
                 candidates = []
         if not candidates:
             candidates = [scope_prefix]
-        self.state.candidate_scopes = candidates[:20]
-        return self.state.candidate_scopes
+        selected_scopes = candidates[:20]
+        self.state.candidate_scopes = selected_scopes
+        return selected_scopes
 
     @listen(filter_and_chunk)
     def search_chunks(self) -> list[Any]:
@@ -368,9 +369,10 @@ class RecallFlow(Flow[RecallState]):
                         )
                     )
         matches.sort(key=lambda m: m.score, reverse=True)
-        self.state.final_results = matches[: self.state.limit]
+        final_results = matches[: self.state.limit]
+        self.state.final_results = final_results
 
         if self.state.evidence_gaps and self.state.final_results:
             self.state.final_results[0].evidence_gaps = list(self.state.evidence_gaps)
 
-        return self.state.final_results
+        return final_results

lib/crewai/src/crewai/telemetry/telemetry.py

@@ -30,7 +30,7 @@ from opentelemetry.sdk.trace.export import (
     BatchSpanProcessor,
     SpanExportResult,
 )
-from opentelemetry.trace import Span
+from opentelemetry.trace import ProxyTracerProvider, Span
 from typing_extensions import Self
 
 from crewai.events.event_bus import crewai_event_bus
@@ -162,6 +162,10 @@ class Telemetry:
         if self.ready and not self.trace_set:
             try:
                 with suppress_warnings():
+                    existing_provider = trace.get_tracer_provider()
+                    if not isinstance(existing_provider, ProxyTracerProvider):
+                        self.trace_set = True
+                        return
                     trace.set_tracer_provider(self.provider)
                     self.trace_set = True
             except Exception as e:

lib/crewai/tests/agents/test_agent.py

@@ -4,6 +4,7 @@ import os
 import threading
 from unittest import mock
 from unittest.mock import MagicMock, patch
+import warnings
 
 from crewai.agents.crew_agent_executor import AgentFinish, CrewAgentExecutor
 from crewai.constants import DEFAULT_LLM_MODEL
@@ -77,6 +78,51 @@ def test_agent_creation():
     assert agent.backstory == "test backstory"
 
 
+def test_agent_exposes_i18n_for_backward_compatibility():
+    from crewai.utilities.i18n import I18N_DEFAULT
+
+    agent = Agent(role="test role", goal="test goal", backstory="test backstory")
+
+    with pytest.warns(DeprecationWarning, match="Agent.i18n is deprecated"):
+        i18n = agent.i18n
+
+    assert i18n is I18N_DEFAULT
+    assert isinstance(i18n.slice("role_playing"), str)
+
+
+def test_agent_accepts_custom_i18n():
+    from crewai.utilities.i18n import I18N
+
+    prompt_file = os.path.join(
+        os.path.dirname(__file__), "..", "utilities", "prompts.json"
+    )
+    i18n = I18N(prompt_file=prompt_file)
+    agent = Agent(
+        role="test role",
+        goal="test goal",
+        backstory="test backstory",
+        i18n=i18n,
+    )
+
+    with pytest.warns(DeprecationWarning, match="Agent.i18n is deprecated"):
+        agent_i18n = agent.i18n
+
+    assert agent_i18n is i18n
+    assert agent_i18n.slice("role_playing") == "Lorem ipsum dolor sit amet"
+
+
+def test_agent_copy_does_not_emit_i18n_deprecation_warning():
+    agent = Agent(role="test role", goal="test goal", backstory="test backstory")
+
+    with warnings.catch_warnings(record=True) as caught_warnings:
+        warnings.simplefilter("always", DeprecationWarning)
+        agent.copy()
+
+    assert not any(
+        "Agent.i18n is deprecated" in str(w.message) for w in caught_warnings
+    )
+
+
 def test_agent_with_only_system_template():
     """Test that an agent with only system_template works without errors."""
     agent = Agent(

lib/crewai/tests/agents/test_agent_executor.py

@@ -32,7 +32,7 @@ def _build_executor(**kwargs: Any) -> AgentExecutor:
     executor._method_outputs = []
     executor._completed_methods = set()
     executor._fired_or_listeners = set()
-    executor._pending_and_listeners = {}
+    executor._pending_events = {}
     executor._method_execution_counts = {}
     executor._method_call_counts = {}
     executor._event_futures = []

lib/crewai/tests/events/test_event_replay.py

@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+import threading
 from typing import Any
 from unittest.mock import patch
 
@@ -109,10 +110,79 @@ class TestCheckpointListenerOptsOut:
             assert do_cp.call_count == 0
 
 
-class TestFlowResumeReplaysEvents:
-    """End-to-end: a resumed flow emits MethodExecution* events for completed methods."""
+class TestCheckpointResumeReplaysEvents:
+    """A flow resumed from a checkpoint replays MethodExecution* events for
+    completed methods and executes the pending ones. The checkpoint persists
+    the event record, which is reloaded into the per-run runtime state.
 
-    def test_resume_dispatches_completed_method_events(self, tmp_path) -> None:
+    ``step_c`` is gated on a threading.Event so the flow is frozen with exactly
+    ``step_a`` and ``step_b`` completed when the checkpoint is written — the
+    mid-run snapshot is deterministic rather than dependent on write timing.
+    """
+
+    def test_resume_replays_completed_and_executes_pending(self, tmp_path) -> None:
+        from crewai.flow.flow import Flow, listen, start
+        from crewai.state.checkpoint_config import CheckpointConfig
+
+        at_step_c = threading.Event()
+        release = threading.Event()
+        captured: list[Any] = []
+
+        class ThreeStepFlow(Flow[dict]):
+            @start()
+            def step_a(self) -> str:
+                return "a"
+
+            @listen(step_a)
+            def step_b(self) -> str:
+                return "b"
+
+            @listen(step_b)
+            def step_c(self) -> str:
+                captured.append(crewai_event_bus.runtime_state)
+                at_step_c.set()
+                release.wait(timeout=10)
+                return "c"
+
+        runner = threading.Thread(target=ThreeStepFlow().kickoff)
+        runner.start()
+        try:
+            assert at_step_c.wait(timeout=10)
+            location = captured[0].checkpoint(str(tmp_path / "cp"))
+        finally:
+            release.set()
+            runner.join(timeout=10)
+
+        captured_started: list[str] = []
+        captured_finished: list[str] = []
+
+        with crewai_event_bus.scoped_handlers():
+
+            @crewai_event_bus.on(MethodExecutionStartedEvent)
+            def _cs(_: Any, event: MethodExecutionStartedEvent) -> None:
+                captured_started.append(event.method_name)
+
+            @crewai_event_bus.on(MethodExecutionFinishedEvent)
+            def _cf(_: Any, event: MethodExecutionFinishedEvent) -> None:
+                captured_finished.append(event.method_name)
+
+            ThreeStepFlow().kickoff(
+                from_checkpoint=CheckpointConfig(restore_from=location)
+            )
+
+        assert captured_started == ["step_a", "step_b", "step_c"]
+        assert captured_finished == ["step_a", "step_b", "step_c"]
+
+
+class TestPersistResumeDoesNotReplayCompletedEvents:
+    """A @persist resume continues from pending methods only.
+
+    @persist stores flow state, not the event record, so completed-method
+    events have no persisted source to replay from. Runtime state is scoped
+    per run, so flow1's events are not visible to flow2.
+    """
+
+    def test_persist_resume_executes_only_pending_methods(self, tmp_path) -> None:
         from crewai.flow.flow import Flow, listen, start
         from crewai.flow.persistence.sqlite import SQLiteFlowPersistence
 
@@ -132,9 +202,6 @@ class TestFlowResumeReplaysEvents:
             def step_c(self) -> str:
                 return "c"
 
-        if crewai_event_bus.runtime_state is not None:
-            crewai_event_bus.runtime_state.event_record.clear()
-
         flow1 = ThreeStepFlow(persistence=persistence)
         flow1.kickoff()
         flow_id = flow1.state["id"]
@@ -157,9 +224,5 @@ class TestFlowResumeReplaysEvents:
 
             flow2.kickoff(inputs={"id": flow_id})
 
-        assert captured_started.count("step_a") == 1
-        assert captured_started.count("step_b") == 1
-        assert captured_started.count("step_c") == 1
-        assert captured_finished.count("step_a") == 1
-        assert captured_finished.count("step_b") == 1
-        assert captured_finished.count("step_c") == 1
+        assert captured_started == ["step_c"]
+        assert captured_finished == ["step_c"]

lib/crewai/tests/telemetry/test_telemetry.py

@@ -6,6 +6,7 @@ import pytest
 from crewai import Agent, Crew, Task
 from crewai.telemetry import Telemetry
 from opentelemetry import trace
+from opentelemetry.sdk.trace import TracerProvider
 
 
 @pytest.fixture(autouse=True)
@@ -53,6 +54,23 @@ def test_telemetry_enabled_by_default():
             assert telemetry.ready is True
 
 
+def test_set_tracer_skips_when_provider_already_configured():
+    """A second telemetry instance must not re-install the global provider."""
+    with (
+        patch.dict(os.environ, {}, clear=True),
+        patch(
+            "crewai.telemetry.telemetry.trace.get_tracer_provider",
+            return_value=TracerProvider(),
+        ),
+        patch("crewai.telemetry.telemetry.trace.set_tracer_provider") as mock_set,
+    ):
+        telemetry = Telemetry()
+        telemetry.set_tracer()
+
+    mock_set.assert_not_called()
+    assert telemetry.trace_set is True
+
+
 @patch("crewai.telemetry.telemetry.logger.error")
 @patch(
     "opentelemetry.exporter.otlp.proto.http.trace_exporter.OTLPSpanExporter.export",

lib/crewai/tests/test_checkpoint.py

@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+import inspect
 import json
 import os
 import sqlite3
@@ -16,6 +17,7 @@ from pydantic import BaseModel
 from crewai.agent.core import Agent
 from crewai.agents.agent_builder.base_agent import BaseAgent
 from crewai.crew import Crew
+from crewai.llms.base_llm import BaseLLM
 from crewai.flow.flow import _INITIAL_STATE_CLASS_MARKER, Flow, start
 from crewai.state.checkpoint_config import CheckpointConfig
 from crewai.state.checkpoint_listener import (
@@ -682,3 +684,85 @@ class TestAgentCheckpoint:
             cfg = CheckpointConfig(restore_from=loc)
             restored = Agent.from_checkpoint(cfg)
             assert restored._kickoff_event_id == "evt-456"
+
+
+class _FinalAnswerLLM(BaseLLM):
+    """Stub LLM that always returns a final answer without any API calls."""
+
+    def __init__(self) -> None:
+        super().__init__(model="stub")
+
+    def call(
+        self,
+        messages,
+        tools=None,
+        callbacks=None,
+        available_functions=None,
+        from_task=None,
+        from_agent=None,
+        response_model=None,
+    ):
+        return "Final Answer: done."
+
+    def supports_function_calling(self) -> bool:
+        return False
+
+    def supports_stop_words(self) -> bool:
+        return False
+
+    def get_context_window_size(self) -> int:
+        return 4096
+
+    async def acall(self, *args, **kwargs):
+        raise NotImplementedError
+
+
+class TestCheckpointReusedExecutor:
+    """Checkpoint serialization stamps every live Flow's completed methods.
+
+    The agent executor is a Flow reused across a crew's tasks, so the stamp
+    must not be read back as a restore signal on the next task — otherwise the
+    second task replays as a resume and never reaches a final answer.
+    """
+
+    def test_second_task_runs_with_checkpointing_enabled(self) -> None:
+        agent = Agent(role="r", goal="g", backstory="b", llm=_FinalAnswerLLM())
+        task1 = Task(description="first", expected_output="x", agent=agent)
+        task2 = Task(description="second", expected_output="y", agent=agent)
+        with tempfile.TemporaryDirectory() as d:
+            crew = Crew(
+                agents=[agent],
+                tasks=[task1, task2],
+                verbose=False,
+                checkpoint=CheckpointConfig(
+                    provider=JsonProvider(location=d),
+                    on_events=["task_started", "task_completed"],
+                ),
+            )
+            result = crew.kickoff()
+
+        assert len(result.tasks_output) == 2
+        assert result.tasks_output[1].raw
+
+
+class TestCustomLLMCheckpointRestore:
+    """A custom BaseLLM subclass serializes with the inherited llm_type "base".
+
+    Restoring it must not try to instantiate the abstract BaseLLM; it is rebuilt
+    as a concrete LLM from the saved config instead.
+    """
+
+    def test_restore_does_not_instantiate_abstract_base_llm(self) -> None:
+        agent = Agent(role="r", goal="g", backstory="b", llm=_FinalAnswerLLM())
+        task = Task(description="d", expected_output="e", agent=agent)
+        crew = Crew(agents=[agent], tasks=[task], verbose=False)
+
+        raw = RuntimeState(root=[crew]).model_dump_json()
+        restored = RuntimeState.model_validate_json(
+            raw, context={"from_checkpoint": True}
+        )
+
+        llm = restored.root[0].agents[0].llm
+        assert isinstance(llm, BaseLLM)
+        assert not inspect.isabstract(type(llm))
+        assert llm.model == "stub"

lib/crewai/tests/test_event_record.py

@@ -409,4 +409,31 @@ class TestRuntimeStateIntegration:
             old_json, context={"from_checkpoint": True}
         )
         assert len(restored.root) == 1
-        assert len(restored.event_record) == 0
+        assert len(restored.event_record) == 0
+
+    def test_reset_runtime_state_clears_state_and_registry(self):
+        from crewai import Agent, Crew, RuntimeState
+        from crewai.events.event_bus import crewai_event_bus
+
+        if RuntimeState is None:
+            pytest.skip("RuntimeState unavailable (model_rebuild failed)")
+
+        agent = Agent(role="test", goal="test", backstory="test", llm="gpt-4o-mini")
+        crew = Crew(agents=[agent], tasks=[], verbose=False)
+
+        previous_state = crewai_event_bus._runtime_state
+        previous_ids = crewai_event_bus._registered_entity_ids
+        crewai_event_bus._runtime_state = None
+        crewai_event_bus._registered_entity_ids = set()
+        try:
+            crewai_event_bus.register_entity(crew)
+            assert crewai_event_bus.runtime_state is not None
+            assert crewai_event_bus._registered_entity_ids
+
+            crewai_event_bus.reset_runtime_state()
+
+            assert crewai_event_bus.runtime_state is None
+            assert crewai_event_bus._registered_entity_ids == set()
+        finally:
+            crewai_event_bus._runtime_state = previous_state
+            crewai_event_bus._registered_entity_ids = previous_ids

lib/crewai/tests/test_flow.py

@@ -1040,7 +1040,7 @@ def test_flow_plotting():
         received_events.append(event)
         event_received.set()
 
-    flow.plot("test_flow")
+    flow.plot("test_flow", show=False)
 
     assert event_received.wait(timeout=5), "Timeout waiting for plot event"
     assert len(received_events) == 1
@@ -1542,40 +1542,63 @@ def test_deeply_nested_conditions():
 
 
 def test_or_branch_does_not_leave_stale_and_state():
-    """or_() over nested and_() branches must not leave stale pending AND state.
-
-    Regression: evaluating an or_() condition stopped at the first branch that was
-    satisfied, so a later and_() branch that the *same* trigger would have completed
-    never cleared its pending state. On the next cycle that trigger alone then
-    spuriously re-satisfied the whole condition. Both branches share the final
-    event ``x`` here, so the shared trigger that completes branch ``(a AND x)`` also
-    completes branch ``(c AND x)`` and both must be cleared together.
-    """
+    fired = []
 
     class StaleStateFlow(Flow):
         @start()
         def begin(self):
             pass
 
-        @listen(or_(and_("a", "x"), and_("c", "x")))
-        def joined(self):
+        @listen(begin)
+        def a(self):
             pass
 
-    flow = StaleStateFlow()
-    condition = type(flow)._listen_condition("joined")
+        @listen(begin)
+        def c(self):
+            pass
 
-    def fires(trigger):
-        return flow._evaluate_condition(condition, trigger, "joined")
+        @listen(and_(a, c))
+        def x(self):
+            pass
 
-    # First cycle: "a" then "c" arrive, then the shared "x" completes (a AND x).
-    assert fires("a") is False
-    assert fires("c") is False
-    assert fires("x") is True
+        @listen(or_(and_("a", "x"), and_("c", "y")))
+        def joined(self):
+            fired.append("joined")
 
-    # Next cycle: "x" alone must NOT re-satisfy the condition. The "c" from the
-    # previous cycle was consumed when "joined" fired, so neither branch is half
-    # complete and "x" by itself is insufficient.
-    assert fires("x") is False
+        @router(joined)
+        def emit_y(self):
+            return "y"
+
+    StaleStateFlow().kickoff()
+
+    assert fired == ["joined"]
+
+
+def test_and_branch_inside_or_does_not_race():
+    execution_order = []
+
+    class DiamondWithFallbackFlow(Flow):
+        @start()
+        def go(self):
+            execution_order.append("go")
+
+        @listen(go)
+        def a(self):
+            execution_order.append("a")
+
+        @listen(go)
+        def b(self):
+            execution_order.append("b")
+
+        @listen(or_(and_(a, b), "fallback"))
+        def done(self):
+            execution_order.append("done")
+
+    DiamondWithFallbackFlow().kickoff()
+
+    assert "done" in execution_order
+    assert execution_order.index("done") > execution_order.index("a")
+    assert execution_order.index("done") > execution_order.index("b")
 
 
 def test_mixed_sync_async_execution_order():

lib/crewai/tests/test_flow_conversation.py

@@ -26,7 +26,11 @@ from crewai.experimental import (
     RouterConfig,
 )
 from crewai.flow import Flow, ChatState, listen, start
-from crewai.flow.flow_context import current_flow_id, current_flow_name
+from crewai.flow.flow_context import (
+    current_flow_defer_trace_finalization,
+    current_flow_id,
+    current_flow_name,
+)
 from crewai.flow.conversation import (
     append_message,
     get_conversation_messages,
@@ -169,9 +173,6 @@ class TestConversationalFlow:
         )
 
 
-    @pytest.mark.skip(
-        reason="Experimental conversational registry behavior is out of scope for the definition-first start migration."
-    )
     def test_handle_turn_routes_to_listener_and_records_public_result(self) -> None:
         @ConversationConfig(default_intents=["research"], intent_llm="gpt-4o-mini")
         class ResearchFlow(ConversationalFlow):
@@ -595,18 +596,15 @@ class TestConversationalFlow:
         assert result == "legacy-searched"
         assert flow.state.last_intent == "search"
 
-    @pytest.mark.skip(
-        reason="Experimental conversational sequential-start behavior is out of scope for the definition-first start migration."
-    )
     def test_user_start_methods_run_sequentially_before_router_in_conversational_mode(
         self,
     ) -> None:
         """Conversational flows: user ``@start`` methods finish before router fires.
 
         Non-chat flows run ``@start`` methods in parallel via ``asyncio.gather``,
-        which would race with ``conversation_start`` and let the router fire
+        which would race with ``route_conversation`` and let the router fire
         before user setup finished. In conversational mode the framework runs
-        them sequentially, with ``conversation_start`` last.
+        them sequentially, with ``route_conversation`` last.
         """
         order: list[str] = []
 
@@ -649,18 +647,10 @@ class TestConversationalFlow:
         assert "attach_bus" in order  # still fires every turn
         assert "route_turn" in order
 
-    @pytest.mark.skip(
-        reason="Experimental inherited conversational start registration is out of scope for the definition-first start migration."
-    )
-    def test_subclass_can_override_conversation_start_without_redecorating(
+    def test_subclass_can_override_conversation_start_helper(
         self,
     ) -> None:
-        """Overriding an inherited ``@start`` method must not unregister it.
-
-        Before the metaclass fix, subclasses had to re-apply ``@start()`` on
-        every override or the parent's ``conversation_start`` would silently
-        drop out of the start registry — leaving the flow with nothing to fire.
-        """
+        """The compatibility helper remains overridable without adding a Flow node."""
 
         bootstrap_calls: list[str] = []
 
@@ -681,6 +671,38 @@ class TestConversationalFlow:
         flow = BootstrapFlow()
         flow.handle_turn("hi")
 
+        assert bootstrap_calls == ["ran"]
+        assert "conversation_start" not in BootstrapFlow.flow_definition().methods
+        route_definition = BootstrapFlow.flow_definition().methods["route_conversation"]
+        assert route_definition.start is True
+        assert route_definition.router is True
+        assert flow.state.messages[-1].content == "worked"
+
+    def test_legacy_decorated_conversation_start_runs_once_per_turn(
+        self,
+    ) -> None:
+        """Legacy ``@start`` overrides are not invoked again by the router."""
+
+        bootstrap_calls: list[str] = []
+
+        @ConversationConfig()
+        class BootstrapFlow(ConversationalFlow):
+            @start()
+            def conversation_start(self) -> str | None:
+                bootstrap_calls.append("ran")
+                return super().conversation_start()
+
+            def route_turn(self, context: dict[str, Any]) -> str | None:
+                return "work"
+
+            @listen("work")
+            def do_work(self) -> str:
+                self.append_assistant_message("worked")
+                return "worked"
+
+        flow = BootstrapFlow()
+        flow.handle_turn("hi")
+
         assert bootstrap_calls == ["ran"]
         assert flow.state.messages[-1].content == "worked"
 
@@ -1179,6 +1201,40 @@ class TestConversationalFlow:
                     "finalize_session_traces must finalize the trace batch once"
                 )
 
+    def test_deferred_resume_skips_per_resume_flow_finished_event(self) -> None:
+        """Deferred sessions do not emit terminal events while resuming."""
+        from crewai.events.types.flow_events import FlowFinishedEvent
+        from crewai.flow.async_feedback.types import PendingFeedbackContext
+
+        class DeferredResumeFlow(Flow[ChatState]):
+            defer_trace_finalization = True
+
+            @start()
+            def begin(self) -> str:
+                return "started"
+
+        flow = DeferredResumeFlow()
+        flow._pending_feedback_context = PendingFeedbackContext(
+            flow_id=flow.flow_id,
+            flow_class="DeferredResumeFlow",
+            method_name="begin",
+            method_output="started",
+            message="Review",
+        )
+
+        finished_events: list[FlowFinishedEvent] = []
+
+        with crewai_event_bus.scoped_handlers():
+
+            @crewai_event_bus.on(FlowFinishedEvent)
+            def capture(_: Any, event: FlowFinishedEvent) -> None:
+                finished_events.append(event)
+
+            flow.resume("approved")
+            crewai_event_bus.flush()
+
+        assert finished_events == []
+
     def test_finalize_session_traces_restores_event_scope(self, capsys) -> None:
         """No ``empty scope stack`` warning when deferred ``flow_finished`` fires.
 
@@ -1281,7 +1337,11 @@ class TestFlowTracingWhenSuppressed:
 
         assert started == ["QuietFlow"]
 
-    def test_method_execution_emitted_when_panel_events_suppressed(self) -> None:
+    def test_method_execution_suppressed_when_flow_events_suppressed(self) -> None:
+        """``suppress_flow_events=True`` silences MethodExecution events so
+        infrastructure flows (AgentExecutor, memory) don't emit one trace span
+        per internal control-flow method."""
+
         class QuietFlow(Flow[ChatState]):
             suppress_flow_events = True
 
@@ -1303,8 +1363,8 @@ class TestFlowTracingWhenSuppressed:
         with patch.object(crewai_event_bus, "emit", side_effect=track_emit):
             QuietFlow().kickoff()
 
-        assert started == ["begin"]
-        assert finished == ["begin"]
+        assert started == []
+        assert finished == []
 
     def test_llm_action_inside_flow_claims_flow_trace_batch(self) -> None:
         listener = TraceCollectionListener()
@@ -1338,6 +1398,12 @@ class TestFlowTracingWhenSuppressed:
 
 
 class TestDeferTraceFinalization:
+    def test_bare_conversational_flow_defers_by_default(self) -> None:
+        class BareChat(ConversationalFlow):
+            pass
+
+        assert BareChat()._should_defer_trace_finalization() is True
+
     def test_conversation_config_drives_defer_flag(self) -> None:
         """``ConversationConfig(defer_trace_finalization=...)`` controls whether
         a conversational subclass defers per-turn trace finalization."""
@@ -1470,6 +1536,44 @@ class TestDeferredFlowLifecycleEvents:
                     listener.batch_manager.finalize_batch()
             mock_finalize.assert_not_called()
 
+    def test_deferred_flow_kickoff_marks_trace_manager_session_deferred(
+        self,
+    ) -> None:
+        class DeferredTraceFlow(Flow[ChatState]):
+            @start()
+            def begin(self) -> str:
+                return "done"
+
+        listener = TraceCollectionListener()
+        listener.batch_manager.defer_session_finalization = False
+
+        flow = DeferredTraceFlow()
+        flow.defer_trace_finalization = True
+
+        with patch.object(listener.batch_manager, "finalize_batch"):
+            flow.kickoff()
+
+        assert listener.batch_manager.defer_session_finalization is True
+
+        flow.finalize_session_traces()
+
+        assert listener.batch_manager.defer_session_finalization is False
+
+    def test_non_deferred_flow_kickoff_clears_stale_trace_manager_flag(
+        self,
+    ) -> None:
+        class PlainTraceFlow(Flow[ChatState]):
+            @start()
+            def begin(self) -> str:
+                return "done"
+
+        listener = TraceCollectionListener()
+        listener.batch_manager.defer_session_finalization = True
+
+        PlainTraceFlow().kickoff()
+
+        assert listener.batch_manager.defer_session_finalization is False
+
 
 class TestNestedCrewTracing:
     def test_is_inside_active_flow_context_when_kickoff_running(self) -> None:
@@ -1523,3 +1627,130 @@ class TestNestedCrewTracing:
             elif listener.batch_manager.batch_owner_type == "crew":
                 listener.batch_manager.finalize_batch()
             mock_finalize.assert_not_called()
+
+    def test_lazy_flow_batch_from_context_preserves_deferred_parent(self) -> None:
+        from crewai.events.listeners.tracing.trace_listener import (
+            TraceCollectionListener,
+        )
+
+        listener = TraceCollectionListener()
+        listener.batch_manager.current_batch = None
+        listener.batch_manager.batch_owner_type = None
+        listener.batch_manager.batch_owner_id = None
+        listener.batch_manager.defer_session_finalization = False
+        listener.batch_manager.event_buffer.clear()
+
+        flow_id_token = current_flow_id.set("parent-flow-id")
+        flow_name_token = current_flow_name.set("ParentChatFlow")
+        defer_token = current_flow_defer_trace_finalization.set(True)
+        try:
+            initialized = listener._try_initialize_flow_batch_from_context(
+                type("Event", (), {"timestamp": None})()
+            )
+
+            assert initialized is True
+            assert listener.batch_manager.batch_owner_type == "flow"
+            assert listener.batch_manager.batch_owner_id == "parent-flow-id"
+            assert listener.batch_manager.defer_session_finalization is True
+            assert listener.batch_manager.current_batch is not None
+            assert (
+                listener.batch_manager.current_batch.execution_metadata[
+                    "execution_type"
+                ]
+                == "flow"
+            )
+            assert (
+                listener.batch_manager.current_batch.execution_metadata["flow_name"]
+                == "ParentChatFlow"
+            )
+        finally:
+            current_flow_defer_trace_finalization.reset(defer_token)
+            current_flow_name.reset(flow_name_token)
+            current_flow_id.reset(flow_id_token)
+            listener.batch_manager.current_batch = None
+            listener.batch_manager.batch_owner_type = None
+            listener.batch_manager.batch_owner_id = None
+            listener.batch_manager.trace_batch_id = None
+            listener.batch_manager.defer_session_finalization = False
+            listener.batch_manager.event_buffer.clear()
+
+    def test_nested_agent_executor_flow_does_not_finalize_parent_batch(
+        self,
+    ) -> None:
+        from crewai import Agent, Crew, Task
+        from crewai.llms.base_llm import BaseLLM
+
+        class StaticLLM(BaseLLM):
+            def __init__(self) -> None:
+                super().__init__(model="debug-static-llm", provider="debug")
+
+            def call(
+                self,
+                messages: Any,
+                tools: Any = None,
+                callbacks: Any = None,
+                available_functions: Any = None,
+                from_task: Any = None,
+                from_agent: Any = None,
+                response_model: Any = None,
+            ) -> str:
+                return (
+                    "Thought: I can answer directly.\n"
+                    "Final Answer: nested crew result"
+                )
+
+        class NestedCrewFlow(Flow[ChatState]):
+            defer_trace_finalization = True
+            tracing = True
+
+            @start()
+            def begin(self) -> str:
+                return "run_nested_crew"
+
+            @listen(begin)
+            def run_nested_crew(self, _: str) -> str:
+                agent = Agent(
+                    role="Debug Agent",
+                    goal="Return a short deterministic result",
+                    backstory="Used only for trace finalization debugging.",
+                    llm=StaticLLM(),
+                    verbose=False,
+                )
+                task = Task(
+                    description="Return the deterministic nested crew result.",
+                    expected_output="nested crew result",
+                    agent=agent,
+                )
+                return Crew(agents=[agent], tasks=[task], verbose=False).kickoff().raw
+
+        listener = TraceCollectionListener()
+        listener.batch_manager.current_batch = None
+        listener.batch_manager.batch_owner_type = None
+        listener.batch_manager.batch_owner_id = None
+        listener.batch_manager.trace_batch_id = None
+        listener.batch_manager.defer_session_finalization = False
+        listener.batch_manager.event_buffer.clear()
+        listener.first_time_handler.is_first_time = False
+
+        def initialize_backend_batch(*_: Any, **__: Any) -> None:
+            listener.batch_manager.trace_batch_id = "debug-trace-batch"
+
+        flow = NestedCrewFlow()
+
+        with (
+            patch.object(
+                listener.batch_manager,
+                "_initialize_backend_batch",
+                side_effect=initialize_backend_batch,
+            ),
+            patch.object(listener.batch_manager, "finalize_batch") as mock_finalize,
+        ):
+            flow.kickoff()
+            crewai_event_bus.flush()
+            flow.kickoff()
+            crewai_event_bus.flush()
+
+            assert mock_finalize.call_count == 0, (
+                "nested AgentExecutor flows inside a deferred parent Flow must "
+                "not finalize the parent trace batch"
+            )

lib/crewai/tests/test_flow_definition.py

@@ -13,6 +13,7 @@ from pydantic import BaseModel
 import crewai.flow.dsl as flow_dsl
 import crewai.flow.flow_definition as flow_definition
 import crewai.flow.visualization.builder as visualization_builder
+from crewai.experimental import ConversationConfig, RouterConfig
 from crewai.flow import Flow, and_, human_feedback, listen, or_, persist, router, start
 
 
@@ -36,6 +37,8 @@ def test_flow_public_exports_are_explicit():
     }
     assert set(flow_definition.__all__) == {
         "FlowConfigDefinition",
+        "FlowConversationalDefinition",
+        "FlowConversationalRouterDefinition",
         "FlowDefinition",
         "FlowDefinitionCondition",
         "FlowDefinitionDiagnostic",
@@ -169,6 +172,7 @@ def test_flow_definition_maps_dsl_to_static_contract():
     assert definition.state.ref and "ContractState" in definition.state.ref
     assert definition.config.stream is True
     assert definition.config.max_method_calls == 7
+    assert definition.conversational is None
 
     assert definition.methods["begin"].start is True
     assert definition.methods["process"].listen == "begin"
@@ -201,25 +205,75 @@ def test_flow_definition_excludes_conversational_builtins_for_regular_flows():
 
     methods = RegularFlow.flow_definition().methods
 
+    assert RegularFlow.flow_definition().conversational is None
     assert set(methods) == {"begin"}
     assert "conversation_start" not in methods
     assert "route_conversation" not in methods
     assert "converse_turn" not in methods
 
 
-@pytest.mark.skip(
-    reason="Experimental conversational inherited built-ins are out of scope for the definition-first start migration."
-)
 def test_flow_definition_includes_conversational_builtins_when_enabled():
     class ChatFlow(Flow):
         conversational = True
 
-    methods = ChatFlow.flow_definition().methods
+    definition = ChatFlow.flow_definition()
+    methods = definition.methods
 
-    assert "conversation_start" in methods
+    assert definition.conversational is not None
+    assert definition.conversational.enabled is True
+    assert definition.conversational.defer_trace_finalization is True
+    assert definition.conversational.builtin_routes == ["converse", "end"]
+    assert "conversation_start" not in methods
     assert "route_conversation" in methods
     assert "converse_turn" in methods
-    assert methods["conversation_start"].start is True
+    assert methods["route_conversation"].start is True
+    assert methods["route_conversation"].router is True
+
+
+def test_flow_definition_serializes_conversational_config():
+    @ConversationConfig(
+        system_prompt="Be concise.",
+        llm="gpt-4o-mini",
+        router=RouterConfig(
+            prompt="Pick a route.",
+            routes=["research"],
+            default_intent="converse",
+            fallback_intent="end",
+        ),
+        default_intents=["research"],
+        visible_agent_outputs=["researcher"],
+        defer_trace_finalization=False,
+    )
+    class ChatFlow(Flow):
+        conversational = True
+
+    conversational = ChatFlow.flow_definition().conversational
+
+    assert conversational is not None
+    assert conversational.system_prompt == "Be concise."
+    assert conversational.llm == "gpt-4o-mini"
+    assert conversational.default_intents == ["research"]
+    assert conversational.visible_agent_outputs == ["researcher"]
+    assert conversational.defer_trace_finalization is False
+    assert conversational.router is not None
+    assert conversational.router.prompt == "Pick a route."
+    assert conversational.router.routes == ["research"]
+    assert conversational.router.fallback_intent == "end"
+
+
+def test_flow_definition_uses_collapsed_conversational_router_start():
+    class ChatFlow(Flow):
+        conversational = True
+
+        def conversation_start(self) -> str | None:
+            return "custom"
+
+    methods = ChatFlow.flow_definition().methods
+
+    assert "conversation_start" not in methods
+    assert "route_conversation" in methods
+    assert methods["route_conversation"].start is True
+    assert methods["route_conversation"].router is True
 
 
 def test_flow_definition_serializes_human_feedback_metadata():

lib/crewai/tests/test_flow_usage_metrics.py

@@ -0,0 +1,394 @@
+"""Tests for flow-level token usage aggregation
+
+``flow.usage_metrics`` listens to ``LLMCallCompletedEvent`` for the duration
+of ``kickoff_async`` so it covers every LLM call inside the flow — crew-led,
+tool-led, AND bare ``LLM.call(...)`` from a flow method. We exercise the
+aggregator end-to-end through the real event bus with fabricated events and
+explicit contextvar control; no live LLM provider is required.
+"""
+
+from __future__ import annotations
+
+import contextvars
+import os
+import tempfile
+from typing import Any, Callable
+from uuid import uuid4
+
+import pytest
+
+from crewai.events.event_bus import crewai_event_bus
+from crewai.events.types.llm_events import LLMCallCompletedEvent, LLMCallType
+from crewai.flow.async_feedback.types import PendingFeedbackContext
+from crewai.flow.flow import Flow, listen, start
+from crewai.flow.flow_context import current_flow_id
+from crewai.flow.persistence.sqlite import SQLiteFlowPersistence
+from crewai.flow.runtime import _usage_dict_to_metrics
+from crewai.types.usage_metrics import UsageMetrics
+
+
+def _emit_llm_call(
+    *,
+    flow_id: str | None,
+    prompt_tokens: int = 0,
+    completion_tokens: int = 0,
+    cached_prompt_tokens: int = 0,
+    reasoning_tokens: int = 0,
+    cache_creation_tokens: int = 0,
+) -> None:
+    """Emit one fake ``LLMCallCompletedEvent`` with ``current_flow_id`` pinned
+    to ``flow_id``.
+
+    Runs in a freshly-copied context so the value the bus snapshots at emit
+    time is exactly ``flow_id`` — independent of the calling thread's outer
+    context. Mirrors how the real ``LLM.call`` emits events at runtime.
+    """
+    usage: dict[str, Any] = {
+        "prompt_tokens": prompt_tokens,
+        "completion_tokens": completion_tokens,
+        "total_tokens": prompt_tokens + completion_tokens,
+    }
+    for key, value in (
+        ("cached_prompt_tokens", cached_prompt_tokens),
+        ("reasoning_tokens", reasoning_tokens),
+        ("cache_creation_tokens", cache_creation_tokens),
+    ):
+        if value:
+            usage[key] = value
+    event = LLMCallCompletedEvent(
+        call_id=str(uuid4()),
+        model="gpt-4o-mini",
+        response="ok",
+        call_type=LLMCallType.LLM_CALL,
+        usage=usage,
+    )
+
+    ctx = contextvars.copy_context()
+
+    def _emit() -> None:
+        current_flow_id.set(flow_id)
+        future = crewai_event_bus.emit(object(), event)
+        if future is not None:
+            future.result(timeout=5.0)
+
+    ctx.run(_emit)
+
+
+class _ScriptedFlow(Flow):
+    """A Flow whose ``@start`` delegates to a per-instance ``_script`` closure.
+
+    Each test attaches a script with ``flow._script = lambda f: ...`` so we
+    don't redefine a Flow subclass for every scenario.
+    """
+
+    @start()
+    def run(self) -> None:
+        script: Callable[[Flow], None] = getattr(self, "_script", lambda _f: None)
+        script(self)
+
+
+def _run(script: Callable[[Flow], None] = lambda _f: None) -> Flow:
+    """Build a ``_ScriptedFlow``, attach ``script``, kickoff. Returns the flow."""
+    flow = _ScriptedFlow()
+    flow._script = script
+    flow.kickoff()
+    return flow
+
+
+class TestUsageDictToMetrics:
+    """Unit tests for the dict-to-UsageMetrics normalizer."""
+
+    @pytest.mark.parametrize(
+        "usage, expected",
+        [
+            (None, None),
+            ({}, None),
+            (
+                {"prompt_tokens": 10, "completion_tokens": 20, "total_tokens": 30},
+                UsageMetrics(
+                    prompt_tokens=10,
+                    completion_tokens=20,
+                    total_tokens=30,
+                    successful_requests=1,
+                ),
+            ),
+            # total_tokens missing → derived from prompt + completion
+            (
+                {"prompt_tokens": 4, "completion_tokens": 6},
+                UsageMetrics(
+                    prompt_tokens=4,
+                    completion_tokens=6,
+                    total_tokens=10,
+                    successful_requests=1,
+                ),
+            ),
+            # Extended provider-specific keys flow through normalization
+            (
+                {
+                    "prompt_tokens": 100,
+                    "completion_tokens": 80,
+                    "total_tokens": 180,
+                    "cached_prompt_tokens": 40,
+                    "reasoning_tokens": 25,
+                    "cache_creation_tokens": 10,
+                },
+                UsageMetrics(
+                    prompt_tokens=100,
+                    completion_tokens=80,
+                    total_tokens=180,
+                    cached_prompt_tokens=40,
+                    reasoning_tokens=25,
+                    cache_creation_tokens=10,
+                    successful_requests=1,
+                ),
+            ),
+            # Garbage / non-int values coerce to 0 instead of crashing
+            (
+                {"prompt_tokens": "n/a", "completion_tokens": None, "total_tokens": 7},
+                UsageMetrics(
+                    prompt_tokens=0,
+                    completion_tokens=0,
+                    total_tokens=7,
+                    successful_requests=1,
+                ),
+            ),
+        ],
+        ids=["none", "empty", "all_keys", "no_total", "extended_keys", "garbage"],
+    )
+    def test_normalization(
+        self, usage: dict[str, Any] | None, expected: UsageMetrics | None
+    ) -> None:
+        assert _usage_dict_to_metrics(usage) == expected
+
+
+class TestFlowUsageAggregation:
+    """End-to-end tests driving the listener through the real event bus."""
+
+    def test_sums_every_llm_call_in_the_flow(self) -> None:
+        """Multiple LLM calls — including bare ``LLM.call(...)`` made outside
+        any crew — accumulate; ``successful_requests`` tracks the call count."""
+
+        def script(flow: Flow) -> None:
+            _emit_llm_call(flow_id=flow._flow_match_id, prompt_tokens=300, completion_tokens=300)
+            _emit_llm_call(flow_id=flow._flow_match_id, prompt_tokens=200, completion_tokens=100)
+            _emit_llm_call(flow_id=flow._flow_match_id, prompt_tokens=20, completion_tokens=20)
+
+        flow = _run(script)
+
+        assert flow.usage_metrics.total_tokens == 940
+        assert flow.usage_metrics.prompt_tokens == 520
+        assert flow.usage_metrics.completion_tokens == 420
+        assert flow.usage_metrics.successful_requests == 3
+
+    def test_returns_zero_when_no_calls_happen(self) -> None:
+        flow = _run()
+        assert flow.usage_metrics == UsageMetrics()
+
+    def test_ignores_events_from_other_flows(self) -> None:
+        """Concurrent flow runs share the singleton bus, so the listener must
+        scope itself to its own flow via the contextvar match."""
+
+        def script(flow: Flow) -> None:
+            _emit_llm_call(flow_id=flow._flow_match_id, prompt_tokens=50, completion_tokens=50)
+            _emit_llm_call(flow_id="some-other-flow", prompt_tokens=49_000, completion_tokens=50_999)
+
+        flow = _run(script)
+
+        assert flow.usage_metrics.total_tokens == 100
+        assert flow.usage_metrics.successful_requests == 1
+
+    def test_resets_between_kickoffs(self) -> None:
+        flow = _ScriptedFlow()
+        flow._script = lambda f: _emit_llm_call(
+            flow_id=f._flow_match_id, prompt_tokens=250, completion_tokens=250
+        )
+
+        flow.kickoff()
+        flow.kickoff()
+
+        assert flow.usage_metrics.total_tokens == 500
+        assert flow.usage_metrics.successful_requests == 1
+
+    def test_usage_metrics_returns_independent_copy(self) -> None:
+        """``usage_metrics`` must return a copy, not the internal instance —
+        otherwise callers can clobber the in-flight accumulator."""
+
+        flow = _run(
+            lambda f: _emit_llm_call(
+                flow_id=f._flow_match_id, prompt_tokens=50, completion_tokens=50
+            )
+        )
+
+        snapshot = flow.usage_metrics
+        snapshot.total_tokens = 999_999
+
+        assert flow.usage_metrics.total_tokens == 100
+
+    def test_handler_is_unregistered_after_kickoff(self) -> None:
+        """Long-lived workers (Celery, devkit) must not leak one handler per
+        kickoff on the singleton bus, on either the success or failure path."""
+
+        def handler_count() -> int:
+            return len(
+                crewai_event_bus._sync_handlers.get(LLMCallCompletedEvent, frozenset())
+            )
+
+        before = handler_count()
+
+        flow = _ScriptedFlow()
+        flow._script = lambda f: _emit_llm_call(
+            flow_id=f._flow_match_id, prompt_tokens=5, completion_tokens=5
+        )
+        for _ in range(3):
+            flow.kickoff()
+
+        assert handler_count() == before
+
+        def boom(_f: Flow) -> None:
+            raise RuntimeError("boom")
+
+        failing = _ScriptedFlow()
+        failing._script = boom
+
+        with pytest.raises(RuntimeError, match="boom"):
+            failing.kickoff()
+
+        assert handler_count() == before
+
+    def test_stale_handler_from_prior_kickoff_does_not_contaminate(self) -> None:
+        """The bus dispatches sync handlers on a thread pool that ``emit``
+        does not wait on. A handler still queued from a prior kickoff
+        must not write into a later kickoff's accumulator — the epoch
+        snapshot in the handler closure bails out on mismatch."""
+
+        captured: dict[str, Any] = {}
+
+        def script(flow: Flow) -> None:
+            _emit_llm_call(flow_id=flow._flow_match_id, prompt_tokens=10, completion_tokens=10)
+            captured["handler"] = flow._usage_aggregation_handler
+            captured["match_id"] = flow._flow_match_id
+
+        flow = _run(script)
+        first_total = flow.usage_metrics.total_tokens
+        assert first_total == 20
+
+        # A second kickoff bumps the epoch and resets the accumulator.
+        flow._script = lambda f: None
+        flow.kickoff()
+        assert flow.usage_metrics.total_tokens == 0
+
+        stale_handler = captured["handler"]
+        assert stale_handler is not None
+
+        stale_event = LLMCallCompletedEvent(
+            call_id=str(uuid4()),
+            model="gpt-4o-mini",
+            response="ok",
+            call_type=LLMCallType.LLM_CALL,
+            usage={"prompt_tokens": 999, "completion_tokens": 999, "total_tokens": 1998},
+        )
+        ctx = contextvars.copy_context()
+        ctx.run(lambda: (current_flow_id.set(captured["match_id"]), stale_handler(object(), stale_event)))
+
+        # Stale handler bailed: second kickoff's accumulator is still zero.
+        assert flow.usage_metrics.total_tokens == 0
+
+    def test_pause_detaches_listener_and_does_not_leak(self) -> None:
+        """When ``kickoff_async`` pauses for human feedback, the listener
+        must be detached from the singleton bus to avoid leaking handlers
+        across abandoned paused instances. Pre-pause LLM events still
+        count because the bus snapshots handlers at emit time. Late
+        events emitted after the pause returns do not count for this
+        instance — resume paths re-attach a fresh listener."""
+
+        from crewai.flow.async_feedback.types import HumanFeedbackPending
+
+        captured: dict[str, Any] = {}
+
+        class _PausingFlow(Flow):
+            @start()
+            def begin(self) -> None:
+                _emit_llm_call(
+                    flow_id=self._flow_match_id,
+                    prompt_tokens=10,
+                    completion_tokens=20,
+                )
+                captured["pre_pause_total"] = self.usage_metrics.total_tokens
+                raise HumanFeedbackPending(
+                    context=PendingFeedbackContext(
+                        flow_id=self.flow_id,
+                        flow_class="_PausingFlow",
+                        method_name="begin",
+                        method_output="content",
+                        message="Review:",
+                    )
+                )
+
+        with tempfile.TemporaryDirectory() as tmpdir:
+            persistence = SQLiteFlowPersistence(os.path.join(tmpdir, "f.db"))
+            flow = _PausingFlow(persistence=persistence)
+            result = flow.kickoff()
+
+            assert isinstance(result, HumanFeedbackPending)
+            assert captured["pre_pause_total"] == 30
+            assert flow._usage_aggregation_handler is None
+
+            # A late event emitted after the pause does not reach the
+            # detached listener, so the running total is unchanged.
+            _emit_llm_call(
+                flow_id=flow._flow_match_id,
+                prompt_tokens=2,
+                completion_tokens=3,
+            )
+            assert flow.usage_metrics.total_tokens == 30
+
+    def test_aggregates_resume_after_from_pending(self) -> None:
+        """A flow restored via ``from_pending`` is a fresh instance with no
+        ``_flow_match_id``; without seeding it, the listener attached in
+        ``resume_async`` either ignores its own LLM calls or absorbs unrelated
+        ones. ``from_pending`` must seed the match id so the resume-phase
+        aggregator counts our own calls and only our own calls."""
+
+        class _ResumeFlow(Flow):
+            @start()
+            def begin(self) -> str:
+                return "content"
+
+            @listen(begin)
+            def on_begin(self, _feedback: Any) -> str:
+                _emit_llm_call(
+                    flow_id=self._flow_match_id,
+                    prompt_tokens=100,
+                    completion_tokens=50,
+                )
+                _emit_llm_call(
+                    flow_id="some-other-flow",
+                    prompt_tokens=9_999,
+                    completion_tokens=9_999,
+                )
+                return "done"
+
+        with tempfile.TemporaryDirectory() as tmpdir:
+            persistence = SQLiteFlowPersistence(os.path.join(tmpdir, "f.db"))
+            flow_id = "usage-resume-test"
+            persistence.save_pending_feedback(
+                flow_uuid=flow_id,
+                context=PendingFeedbackContext(
+                    flow_id=flow_id,
+                    flow_class="_ResumeFlow",
+                    method_name="begin",
+                    method_output="content",
+                    message="Review:",
+                ),
+                state_data={"id": flow_id},
+            )
+
+            flow = _ResumeFlow.from_pending(flow_id, persistence)
+            assert flow._flow_match_id == flow.flow_id
+
+            flow.resume("ok")
+
+            assert flow.usage_metrics.total_tokens == 150
+            assert flow.usage_metrics.prompt_tokens == 100
+            assert flow.usage_metrics.completion_tokens == 50
+            assert flow.usage_metrics.successful_requests == 1

lib/crewai/tests/utilities/test_events.py

@@ -838,6 +838,74 @@ def test_flow_method_execution_finished_includes_serialized_state():
     assert final_output == "final_result"
 
 
+def test_suppress_flow_events_silences_method_lifecycle_events():
+    """``suppress_flow_events=True`` emits no MethodExecution* events on the
+    bus (used by infrastructure flows like AgentExecutor so their control-flow
+    methods don't pollute traces), while default flows still emit them."""
+    captured: list[tuple[str, str]] = []
+
+    class SuppressedFlow(Flow):
+        suppress_flow_events: bool = True
+
+        @start()
+        def begin(self):
+            return "started"
+
+        @listen("begin")
+        def process(self):
+            return "done"
+
+    class ControlFlow(Flow):
+        @start()
+        def begin(self):
+            return "started"
+
+        @listen("begin")
+        def process(self):
+            return "done"
+
+    with crewai_event_bus.scoped_handlers():
+
+        @crewai_event_bus.on(MethodExecutionStartedEvent)
+        def _on_started(source, event):
+            captured.append(("started", type(source).__name__))
+
+        @crewai_event_bus.on(MethodExecutionFinishedEvent)
+        def _on_finished(source, event):
+            captured.append(("finished", type(source).__name__))
+
+        SuppressedFlow().kickoff()
+        wait_for_event_handlers()
+        assert [e for e in captured if e[1] == "SuppressedFlow"] == [], (
+            "suppress_flow_events=True must emit no MethodExecution* events"
+        )
+
+        captured.clear()
+        ControlFlow().kickoff()
+        wait_for_event_handlers()
+        control = [e for e in captured if e[1] == "ControlFlow"]
+        assert ("started", "ControlFlow") in control
+        assert ("finished", "ControlFlow") in control
+
+
+def test_infrastructure_flows_suppress_flow_events_by_default():
+    """Pin the infra flows that must stay silent in traces.
+
+    The gating in ``_execute_method`` only helps if these flows actually set
+    ``suppress_flow_events=True``; without this guard, removing the flag from
+    AgentExecutor would silently bring back the verbose per-method trace spans.
+    """
+    from crewai.experimental.agent_executor import AgentExecutor
+    from crewai.memory.encoding_flow import EncodingFlow
+    from crewai.memory.recall_flow import RecallFlow
+
+    assert AgentExecutor.model_fields["suppress_flow_events"].default is True
+
+    for flow_cls in (EncodingFlow, RecallFlow):
+        flow = flow_cls(storage=None, llm=None, embedder=None)
+        assert flow.suppress_flow_events is True
+
+
 @pytest.mark.vcr()
 def test_llm_emits_call_started_event():
     started_events: list[LLMCallStartedEvent] = []

lib/devtools/src/crewai_devtools/__init__.py

@@ -1,3 +1,3 @@
 """CrewAI development tools."""
 
-__version__ = "1.14.7a4"
+__version__ = "1.14.7"