Brandon/cre 130 pipeline project structure (#1066)

* WIP. Procedure appears to be working well. Working on mocking properly for tests

* All tests are passing now

* rshift working

* Add back in Gui's tool_usage fix

* WIP

* Going to start refactoring for pipeline_output

* Update terminology

* new pipeline flow with traces and usage metrics working. need to add more tests and make sure PipelineOutput behaves likew CrewOutput

* Fix pipelineoutput to look more like crewoutput and taskoutput

* Implemented additional tests for pipeline. One test is failing. Need team support

* Update docs for pipeline

* Update pipeline to properly process input and ouput dictionary

* Update Pipeline docs

* Add back in commentary at top of pipeline file

* Starting to work on router

* Drop router for now. will add in separately

* In the middle of fixing router. A ton of circular dependencies. Moving over to a new design.

* WIP.

* Fix circular dependencies and updated PipelineRouter

* Add in Eduardo feedback. Still need to add in more commentary describing the design decisions for pipeline

* Add developer notes to explain what is going on in pipelines.

* Add doc strings

* Fix missing rag datatype

* WIP. Converting usage metrics from a dict to an object

* Fix tests that were checking usage metrics

* Drop todo

* Fix 1 type error in pipeline

* Update pipeline to use UsageMetric

* Add missing doc string

* WIP.

* Change names

* Rename variables based on joaos feedback

* Fix critical circular dependency issues. Now needing to fix trace issue.

* Tests working now!

* Add more tests which showed underlying issue with traces

* Fix tests

* Remove overly complicated test

* Add router example to docs

* Clean up end of docs

* Clean up docs

* Working on creating Crew templates and pipeline templates

* WIP.

* WIP

* Fix poetry install from templates

* WIP

* Restructure

* changes for lorenze

* more todos

* WIP: create pipelines cli working

* wrapped up router

* ignore mypy src on templates

* ignored signature of copy

* fix all verbose

* rm print statements

* brought back correct folders

* fixes missing folders and then rm print statements

* fixed tests

* fixed broken test

* fixed type checker

* fixed type ignore

* ignore types for templates

* needed

* revert

* exclude only required

* rm type errors on templates

* rm excluding type checks for template files on github action

* fixed missing quotes

---------

Co-authored-by: Brandon Hancock <brandon@brandonhancock.io>
This commit is contained in:
Lorenze Jay
2024-08-09 14:13:29 -07:00
committed by GitHub
parent 6583f31459
commit 62f5b2fb2e
87 changed files with 5435 additions and 571 deletions

View File

@@ -1,6 +1,7 @@
from crewai.agent import Agent
from crewai.crew import Crew
from crewai.pipeline import Pipeline
from crewai.process import Process
from crewai.task import Task
__all__ = ["Agent", "Crew", "Process", "Task"]
__all__ = ["Agent", "Crew", "Process", "Task", "Pipeline"]

View File

@@ -1,4 +1,4 @@
from typing import Any, Dict
from crewai.types.usage_metrics import UsageMetrics
class TokenProcess:
@@ -18,10 +18,10 @@ class TokenProcess:
def sum_successful_requests(self, requests: int):
self.successful_requests = self.successful_requests + requests
def get_summary(self) -> Dict[str, Any]:
return {
"total_tokens": self.total_tokens,
"prompt_tokens": self.prompt_tokens,
"completion_tokens": self.completion_tokens,
"successful_requests": self.successful_requests,
}
def get_summary(self) -> UsageMetrics:
return UsageMetrics(
total_tokens=self.total_tokens,
prompt_tokens=self.prompt_tokens,
completion_tokens=self.completion_tokens,
successful_requests=self.successful_requests,
)

View File

@@ -69,7 +69,7 @@ class CrewAgentExecutor(AgentExecutor, CrewAgentExecutorMixin):
)
intermediate_steps: List[Tuple[AgentAction, str]] = []
# Allowing human input given task setting
if self.task.human_input:
if self.task and self.task.human_input:
self.should_ask_for_human_input = True
# Let's start tracking the number of iterations and time elapsed

View File

@@ -1,11 +1,12 @@
import click
import pkg_resources
from crewai.cli.create_crew import create_crew
from crewai.cli.create_pipeline import create_pipeline
from crewai.memory.storage.kickoff_task_outputs_storage import (
KickoffTaskOutputsSQLiteStorage,
)
from .create_crew import create_crew
from .evaluate_crew import evaluate_crew
from .replay_from_task import replay_task_command
from .reset_memories_command import reset_memories_command
@@ -19,10 +20,19 @@ def crewai():
@crewai.command()
@click.argument("project_name")
def create(project_name):
"""Create a new crew."""
create_crew(project_name)
@click.argument("type", type=click.Choice(["crew", "pipeline"]))
@click.argument("name")
@click.option(
"--router", is_flag=True, help="Create a pipeline with router functionality"
)
def create(type, name, router):
"""Create a new crew or pipeline."""
if type == "crew":
create_crew(name)
elif type == "pipeline":
create_pipeline(name, router)
else:
click.secho("Error: Invalid type. Must be 'crew' or 'pipeline'.", fg="red")
@crewai.command()

View File

@@ -1,25 +1,35 @@
import os
from pathlib import Path
import click
from crewai.cli.utils import copy_template
def create_crew(name):
def create_crew(name, parent_folder=None):
"""Create a new crew."""
folder_name = name.replace(" ", "_").replace("-", "_").lower()
class_name = name.replace("_", " ").replace("-", " ").title().replace(" ", "")
click.secho(f"Creating folder {folder_name}...", fg="green", bold=True)
if parent_folder:
folder_path = Path(parent_folder) / folder_name
else:
folder_path = Path(folder_name)
if not os.path.exists(folder_name):
os.mkdir(folder_name)
os.mkdir(folder_name + "/tests")
os.mkdir(folder_name + "/src")
os.mkdir(folder_name + f"/src/{folder_name}")
os.mkdir(folder_name + f"/src/{folder_name}/tools")
os.mkdir(folder_name + f"/src/{folder_name}/config")
with open(folder_name + "/.env", "w") as file:
file.write("OPENAI_API_KEY=YOUR_API_KEY")
click.secho(
f"Creating {'crew' if parent_folder else 'folder'} {folder_name}...",
fg="green",
bold=True,
)
if not folder_path.exists():
folder_path.mkdir(parents=True)
(folder_path / "tests").mkdir(exist_ok=True)
if not parent_folder:
(folder_path / "src" / folder_name).mkdir(parents=True)
(folder_path / "src" / folder_name / "tools").mkdir(parents=True)
(folder_path / "src" / folder_name / "config").mkdir(parents=True)
with open(folder_path / ".env", "w") as file:
file.write("OPENAI_API_KEY=YOUR_API_KEY")
else:
click.secho(
f"\tFolder {folder_name} already exists. Please choose a different name.",
@@ -28,53 +38,34 @@ def create_crew(name):
return
package_dir = Path(__file__).parent
templates_dir = package_dir / "templates"
templates_dir = package_dir / "templates" / "crew"
# List of template files to copy
root_template_files = [
".gitignore",
"pyproject.toml",
"README.md",
]
root_template_files = (
[".gitignore", "pyproject.toml", "README.md"] if not parent_folder else []
)
tools_template_files = ["tools/custom_tool.py", "tools/__init__.py"]
config_template_files = ["config/agents.yaml", "config/tasks.yaml"]
src_template_files = ["__init__.py", "main.py", "crew.py"]
src_template_files = (
["__init__.py", "main.py", "crew.py"] if not parent_folder else ["crew.py"]
)
for file_name in root_template_files:
src_file = templates_dir / file_name
dst_file = Path(folder_name) / file_name
dst_file = folder_path / file_name
copy_template(src_file, dst_file, name, class_name, folder_name)
src_folder = folder_path / "src" / folder_name if not parent_folder else folder_path
for file_name in src_template_files:
src_file = templates_dir / file_name
dst_file = Path(folder_name) / "src" / folder_name / file_name
dst_file = src_folder / file_name
copy_template(src_file, dst_file, name, class_name, folder_name)
for file_name in tools_template_files:
src_file = templates_dir / file_name
dst_file = Path(folder_name) / "src" / folder_name / file_name
copy_template(src_file, dst_file, name, class_name, folder_name)
for file_name in config_template_files:
src_file = templates_dir / file_name
dst_file = Path(folder_name) / "src" / folder_name / file_name
copy_template(src_file, dst_file, name, class_name, folder_name)
if not parent_folder:
for file_name in tools_template_files + config_template_files:
src_file = templates_dir / file_name
dst_file = src_folder / file_name
copy_template(src_file, dst_file, name, class_name, folder_name)
click.secho(f"Crew {name} created successfully!", fg="green", bold=True)
def copy_template(src, dst, name, class_name, folder_name):
"""Copy a file from src to dst."""
with open(src, "r") as file:
content = file.read()
# Interpolate the content
content = content.replace("{{name}}", name)
content = content.replace("{{crew_name}}", class_name)
content = content.replace("{{folder_name}}", folder_name)
# Write the interpolated content to the new file
with open(dst, "w") as file:
file.write(content)
click.secho(f" - Created {dst}", fg="green")

View File

@@ -0,0 +1,107 @@
import shutil
from pathlib import Path
import click
def create_pipeline(name, router=False):
"""Create a new pipeline project."""
folder_name = name.replace(" ", "_").replace("-", "_").lower()
class_name = name.replace("_", " ").replace("-", " ").title().replace(" ", "")
click.secho(f"Creating pipeline {folder_name}...", fg="green", bold=True)
project_root = Path(folder_name)
if project_root.exists():
click.secho(f"Error: Folder {folder_name} already exists.", fg="red")
return
# Create directory structure
(project_root / "src" / folder_name).mkdir(parents=True)
(project_root / "src" / folder_name / "pipelines").mkdir(parents=True)
(project_root / "src" / folder_name / "crews").mkdir(parents=True)
(project_root / "src" / folder_name / "tools").mkdir(parents=True)
(project_root / "tests").mkdir(exist_ok=True)
# Create .env file
with open(project_root / ".env", "w") as file:
file.write("OPENAI_API_KEY=YOUR_API_KEY")
package_dir = Path(__file__).parent
template_folder = "pipeline_router" if router else "pipeline"
templates_dir = package_dir / "templates" / template_folder
# List of template files to copy
root_template_files = [".gitignore", "pyproject.toml", "README.md"]
src_template_files = ["__init__.py", "main.py"]
tools_template_files = ["tools/__init__.py", "tools/custom_tool.py"]
if router:
crew_folders = [
"classifier_crew",
"normal_crew",
"urgent_crew",
]
pipelines_folders = [
"pipelines/__init__.py",
"pipelines/pipeline_classifier.py",
"pipelines/pipeline_normal.py",
"pipelines/pipeline_urgent.py",
]
else:
crew_folders = [
"research_crew",
"write_linkedin_crew",
"write_x_crew",
]
pipelines_folders = ["pipelines/__init__.py", "pipelines/pipeline.py"]
def process_file(src_file, dst_file):
with open(src_file, "r") as file:
content = file.read()
content = content.replace("{{name}}", name)
content = content.replace("{{crew_name}}", class_name)
content = content.replace("{{folder_name}}", folder_name)
content = content.replace("{{pipeline_name}}", class_name)
with open(dst_file, "w") as file:
file.write(content)
# Copy and process root template files
for file_name in root_template_files:
src_file = templates_dir / file_name
dst_file = project_root / file_name
process_file(src_file, dst_file)
# Copy and process src template files
for file_name in src_template_files:
src_file = templates_dir / file_name
dst_file = project_root / "src" / folder_name / file_name
process_file(src_file, dst_file)
# Copy tools files
for file_name in tools_template_files:
src_file = templates_dir / file_name
dst_file = project_root / "src" / folder_name / file_name
shutil.copy(src_file, dst_file)
# Copy pipelines folders
for file_name in pipelines_folders:
src_file = templates_dir / file_name
dst_file = project_root / "src" / folder_name / file_name
process_file(src_file, dst_file)
# Copy crew folders
for crew_folder in crew_folders:
src_crew_folder = templates_dir / "crews" / crew_folder
dst_crew_folder = project_root / "src" / folder_name / "crews" / crew_folder
if src_crew_folder.exists():
shutil.copytree(src_crew_folder, dst_crew_folder)
else:
click.secho(
f"Warning: Crew folder {crew_folder} not found in template.",
fg="yellow",
)
click.secho(f"Pipeline {name} created successfully!", fg="green", bold=True)

View File

@@ -0,0 +1,2 @@
.env
__pycache__/

View File

@@ -0,0 +1,57 @@
# {{crew_name}} Crew
Welcome to the {{crew_name}} Crew project, powered by [crewAI](https://crewai.com). This template is designed to help you set up a multi-agent AI system with ease, leveraging the powerful and flexible framework provided by crewAI. Our goal is to enable your agents to collaborate effectively on complex tasks, maximizing their collective intelligence and capabilities.
## Installation
Ensure you have Python >=3.10 <=3.13 installed on your system. This project uses [Poetry](https://python-poetry.org/) for dependency management and package handling, offering a seamless setup and execution experience.
First, if you haven't already, install Poetry:
```bash
pip install poetry
```
Next, navigate to your project directory and install the dependencies:
1. First lock the dependencies and then install them:
```bash
poetry lock
```
```bash
poetry install
```
### Customizing
**Add your `OPENAI_API_KEY` into the `.env` file**
- Modify `src/{{folder_name}}/config/agents.yaml` to define your agents
- Modify `src/{{folder_name}}/config/tasks.yaml` to define your tasks
- Modify `src/{{folder_name}}/crew.py` to add your own logic, tools and specific args
- Modify `src/{{folder_name}}/main.py` to add custom inputs for your agents and tasks
## Running the Project
To kickstart your crew of AI agents and begin task execution, run this from the root folder of your project:
```bash
poetry run {{folder_name}}
```
This command initializes the {{name}} Crew, assembling the agents and assigning them tasks as defined in your configuration.
This example, unmodified, will run the create a `report.md` file with the output of a research on LLMs in the root folder.
## Understanding Your Crew
The {{name}} Crew is composed of multiple AI agents, each with unique roles, goals, and tools. These agents collaborate on a series of tasks, defined in `config/tasks.yaml`, leveraging their collective skills to achieve complex objectives. The `config/agents.yaml` file outlines the capabilities and configurations of each agent in your crew.
## Support
For support, questions, or feedback regarding the {{crew_name}} Crew or crewAI.
- Visit our [documentation](https://docs.crewai.com)
- Reach out to us through our [GitHub repository](https://github.com/joaomdmoura/crewai)
- [Join our Discord](https://discord.com/invite/X4JWnZnxPb)
- [Chat with our docs](https://chatg.pt/DWjSBZn)
Let's create wonders together with the power and simplicity of crewAI.

View File

@@ -0,0 +1,19 @@
researcher:
role: >
{topic} Senior Data Researcher
goal: >
Uncover cutting-edge developments in {topic}
backstory: >
You're a seasoned researcher with a knack for uncovering the latest
developments in {topic}. Known for your ability to find the most relevant
information and present it in a clear and concise manner.
reporting_analyst:
role: >
{topic} Reporting Analyst
goal: >
Create detailed reports based on {topic} data analysis and research findings
backstory: >
You're a meticulous analyst with a keen eye for detail. You're known for
your ability to turn complex data into clear and concise reports, making
it easy for others to understand and act on the information you provide.

View File

@@ -0,0 +1,16 @@
research_task:
description: >
Conduct a thorough research about {topic}
Make sure you find any interesting and relevant information given
the current year is 2024.
expected_output: >
A list with 10 bullet points of the most relevant information about {topic}
agent: researcher
reporting_task:
description: >
Review the context you got and expand each topic into a full section for a report.
Make sure the report is detailed and contains any and all relevant information.
expected_output: >
A fully fledge reports with a title, mains topics, each with a full section of information.
agent: reporting_analyst

View File

@@ -0,0 +1,58 @@
from pydantic import BaseModel
from crewai import Agent, Crew, Process, Task
from crewai.project import CrewBase, agent, crew, task
# Uncomment the following line to use an example of a custom tool
# from demo_pipeline.tools.custom_tool import MyCustomTool
# Check our tools documentations for more information on how to use them
# from crewai_tools import SerperDevTool
class ResearchReport(BaseModel):
"""Research Report"""
title: str
body: str
@CrewBase
class ResearchCrew():
"""Research Crew"""
agents_config = 'config/agents.yaml'
tasks_config = 'config/tasks.yaml'
@agent
def researcher(self) -> Agent:
return Agent(
config=self.agents_config['researcher'],
verbose=True
)
@agent
def reporting_analyst(self) -> Agent:
return Agent(
config=self.agents_config['reporting_analyst'],
verbose=True
)
@task
def research_task(self) -> Task:
return Task(
config=self.tasks_config['research_task'],
)
@task
def reporting_task(self) -> Task:
return Task(
config=self.tasks_config['reporting_task'],
output_pydantic=ResearchReport
)
@crew
def crew(self) -> Crew:
"""Creates the Research Crew"""
return Crew(
agents=self.agents, # Automatically created by the @agent decorator
tasks=self.tasks, # Automatically created by the @task decorator
process=Process.sequential,
verbose=True,
)

View File

@@ -0,0 +1,51 @@
from crewai import Agent, Crew, Process, Task
from crewai.project import CrewBase, agent, crew, task
# Uncomment the following line to use an example of a custom tool
# from {{folder_name}}.tools.custom_tool import MyCustomTool
# Check our tools documentations for more information on how to use them
# from crewai_tools import SerperDevTool
@CrewBase
class WriteLinkedInCrew():
"""Research Crew"""
agents_config = 'config/agents.yaml'
tasks_config = 'config/tasks.yaml'
@agent
def researcher(self) -> Agent:
return Agent(
config=self.agents_config['researcher'],
verbose=True
)
@agent
def reporting_analyst(self) -> Agent:
return Agent(
config=self.agents_config['reporting_analyst'],
verbose=True
)
@task
def research_task(self) -> Task:
return Task(
config=self.tasks_config['research_task'],
)
@task
def reporting_task(self) -> Task:
return Task(
config=self.tasks_config['reporting_task'],
output_file='report.md'
)
@crew
def crew(self) -> Crew:
"""Creates the {{crew_name}} crew"""
return Crew(
agents=self.agents, # Automatically created by the @agent decorator
tasks=self.tasks, # Automatically created by the @task decorator
process=Process.sequential,
verbose=True,
)

View File

@@ -0,0 +1,14 @@
x_writer_agent:
role: >
Expert Social Media Content Creator specializing in short form written content
goal: >
Create viral-worthy, engaging short form posts that distill complex {topic} information
into compelling 280-character messages
backstory: >
You're a social media virtuoso with a particular talent for short form content. Your posts
consistently go viral due to your ability to craft hooks that stop users mid-scroll.
You've studied the techniques of social media masters like Justin Welsh, Dickie Bush,
Nicolas Cole, and Shaan Puri, incorporating their best practices into your own unique style.
Your superpower is taking intricate {topic} concepts and transforming them into
bite-sized, shareable content that resonates with a wide audience. You know exactly
how to structure a post for maximum impact and engagement.

View File

@@ -0,0 +1,22 @@
write_x_task:
description: >
Using the research report provided, create an engaging short form post about {topic}.
Your post should have a great hook, summarize key points, and be structured for easy
consumption on a digital platform. The post must be under 280 characters.
Follow these guidelines:
1. Start with an attention-grabbing hook
2. Condense the main insights from the research
3. Use clear, concise language
4. Include a call-to-action or thought-provoking question if space allows
5. Ensure the post flows well and is easy to read quickly
Here is the title of the research report you will be using
Title: {title}
Research:
{body}
expected_output: >
A compelling X post under 280 characters that effectively summarizes the key findings
about {topic}, starts with a strong hook, and is optimized for engagement on the platform.
agent: x_writer_agent

View File

@@ -0,0 +1,36 @@
from crewai import Agent, Crew, Process, Task
from crewai.project import CrewBase, agent, crew, task
# Uncomment the following line to use an example of a custom tool
# from demo_pipeline.tools.custom_tool import MyCustomTool
# Check our tools documentations for more information on how to use them
# from crewai_tools import SerperDevTool
@CrewBase
class WriteXCrew:
"""Research Crew"""
agents_config = "config/agents.yaml"
tasks_config = "config/tasks.yaml"
@agent
def x_writer_agent(self) -> Agent:
return Agent(config=self.agents_config["x_writer_agent"], verbose=True)
@task
def write_x_task(self) -> Task:
return Task(
config=self.tasks_config["write_x_task"],
)
@crew
def crew(self) -> Crew:
"""Creates the Write X Crew"""
return Crew(
agents=self.agents, # Automatically created by the @agent decorator
tasks=self.tasks, # Automatically created by the @task decorator
process=Process.sequential,
verbose=True,
)

View File

@@ -0,0 +1,26 @@
#!/usr/bin/env python
import asyncio
from {{folder_name}}.pipelines.pipeline import {{pipeline_name}}Pipeline
async def run():
"""
Run the pipeline.
"""
inputs = [
{"topic": "AI wearables"},
]
pipeline = {{pipeline_name}}Pipeline()
results = await pipeline.kickoff(inputs)
# Process and print results
for result in results:
print(f"Raw output: {result.raw}")
if result.json_dict:
print(f"JSON output: {result.json_dict}")
print("\n")
def main():
asyncio.run(run())
if __name__ == "__main__":
main()

View File

@@ -0,0 +1,87 @@
"""
This pipeline file includes two different examples to demonstrate the flexibility of crewAI pipelines.
Example 1: Two-Stage Pipeline
-----------------------------
This pipeline consists of two crews:
1. ResearchCrew: Performs research on a given topic.
2. WriteXCrew: Generates an X (Twitter) post based on the research findings.
Key features:
- The ResearchCrew's final task uses output_json to store all research findings in a JSON object.
- This JSON object is then passed to the WriteXCrew, where tasks can access the research findings.
Example 2: Two-Stage Pipeline with Parallel Execution
-------------------------------------------------------
This pipeline consists of three crews:
1. ResearchCrew: Performs research on a given topic.
2. WriteXCrew and WriteLinkedInCrew: Run in parallel, using the research findings to generate posts for X and LinkedIn, respectively.
Key features:
- Demonstrates the ability to run multiple crews in parallel.
- Shows how to structure a pipeline with both sequential and parallel stages.
Usage:
- To switch between examples, comment/uncomment the respective code blocks below.
- Ensure that you have implemented all necessary crew classes (ResearchCrew, WriteXCrew, WriteLinkedInCrew) before running.
"""
# Common imports for both examples
from crewai import Pipeline
# Uncomment the crews you need for your chosen example
from ..crews.research_crew.research_crew import ResearchCrew
from ..crews.write_x_crew.write_x_crew import WriteXCrew
# from .crews.write_linkedin_crew.write_linkedin_crew import WriteLinkedInCrew # Uncomment for Example 2
# EXAMPLE 1: Two-Stage Pipeline
# -----------------------------
# Uncomment the following code block to use Example 1
class {{pipeline_name}}Pipeline:
def __init__(self):
# Initialize crews
self.research_crew = ResearchCrew().crew()
self.write_x_crew = WriteXCrew().crew()
def create_pipeline(self):
return Pipeline(
stages=[
self.research_crew,
self.write_x_crew
]
)
async def kickoff(self, inputs):
pipeline = self.create_pipeline()
results = await pipeline.kickoff(inputs)
return results
# EXAMPLE 2: Two-Stage Pipeline with Parallel Execution
# -------------------------------------------------------
# Uncomment the following code block to use Example 2
# @PipelineBase
# class {{pipeline_name}}Pipeline:
# def __init__(self):
# # Initialize crews
# self.research_crew = ResearchCrew().crew()
# self.write_x_crew = WriteXCrew().crew()
# self.write_linkedin_crew = WriteLinkedInCrew().crew()
# @pipeline
# def create_pipeline(self):
# return Pipeline(
# stages=[
# self.research_crew,
# [self.write_x_crew, self.write_linkedin_crew] # Parallel execution
# ]
# )
# async def run(self, inputs):
# pipeline = self.create_pipeline()
# results = await pipeline.kickoff(inputs)
# return results

View File

@@ -0,0 +1,17 @@
[tool.poetry]
name = "{{folder_name}}"
version = "0.1.0"
description = "{{name}} using crewAI"
authors = ["Your Name <you@example.com>"]
[tool.poetry.dependencies]
python = ">=3.10,<=3.13"
crewai = { extras = ["tools"], version = "^0.46.0" }
asyncio = "*"
[tool.poetry.scripts]
{{folder_name}} = "{{folder_name}}.main:main"
[build-system]
requires = ["poetry-core"]
build-backend = "poetry.core.masonry.api"

View File

@@ -0,0 +1,12 @@
from crewai_tools import BaseTool
class MyCustomTool(BaseTool):
name: str = "Name of my tool"
description: str = (
"Clear description for what this tool is useful for, you agent will need this information to use it."
)
def _run(self, argument: str) -> str:
# Implementation goes here
return "this is an example of a tool output, ignore it and move along."

View File

@@ -0,0 +1,2 @@
.env
__pycache__/

View File

@@ -0,0 +1,57 @@
# {{crew_name}} Crew
Welcome to the {{crew_name}} Crew project, powered by [crewAI](https://crewai.com). This template is designed to help you set up a multi-agent AI system with ease, leveraging the powerful and flexible framework provided by crewAI. Our goal is to enable your agents to collaborate effectively on complex tasks, maximizing their collective intelligence and capabilities.
## Installation
Ensure you have Python >=3.10 <=3.13 installed on your system. This project uses [Poetry](https://python-poetry.org/) for dependency management and package handling, offering a seamless setup and execution experience.
First, if you haven't already, install Poetry:
```bash
pip install poetry
```
Next, navigate to your project directory and install the dependencies:
1. First lock the dependencies and then install them:
```bash
poetry lock
```
```bash
poetry install
```
### Customizing
**Add your `OPENAI_API_KEY` into the `.env` file**
- Modify `src/{{folder_name}}/config/agents.yaml` to define your agents
- Modify `src/{{folder_name}}/config/tasks.yaml` to define your tasks
- Modify `src/{{folder_name}}/crew.py` to add your own logic, tools and specific args
- Modify `src/{{folder_name}}/main.py` to add custom inputs for your agents and tasks
## Running the Project
To kickstart your crew of AI agents and begin task execution, run this from the root folder of your project:
```bash
poetry run {{folder_name}}
```
This command initializes the {{name}} Crew, assembling the agents and assigning them tasks as defined in your configuration.
This example, unmodified, will run the create a `report.md` file with the output of a research on LLMs in the root folder.
## Understanding Your Crew
The {{name}} Crew is composed of multiple AI agents, each with unique roles, goals, and tools. These agents collaborate on a series of tasks, defined in `config/tasks.yaml`, leveraging their collective skills to achieve complex objectives. The `config/agents.yaml` file outlines the capabilities and configurations of each agent in your crew.
## Support
For support, questions, or feedback regarding the {{crew_name}} Crew or crewAI.
- Visit our [documentation](https://docs.crewai.com)
- Reach out to us through our [GitHub repository](https://github.com/joaomdmoura/crewai)
- [Join our Discord](https://discord.com/invite/X4JWnZnxPb)
- [Chat with our docs](https://chatg.pt/DWjSBZn)
Let's create wonders together with the power and simplicity of crewAI.

View File

@@ -0,0 +1,19 @@
researcher:
role: >
{topic} Senior Data Researcher
goal: >
Uncover cutting-edge developments in {topic}
backstory: >
You're a seasoned researcher with a knack for uncovering the latest
developments in {topic}. Known for your ability to find the most relevant
information and present it in a clear and concise manner.
reporting_analyst:
role: >
{topic} Reporting Analyst
goal: >
Create detailed reports based on {topic} data analysis and research findings
backstory: >
You're a meticulous analyst with a keen eye for detail. You're known for
your ability to turn complex data into clear and concise reports, making
it easy for others to understand and act on the information you provide.

View File

@@ -0,0 +1,17 @@
research_task:
description: >
Conduct a thorough research about {topic}
Make sure you find any interesting and relevant information given
the current year is 2024.
expected_output: >
A list with 10 bullet points of the most relevant information about {topic}
agent: researcher
reporting_task:
description: >
Review the context you got and expand each topic into a full section for a report.
Make sure the report is detailed and contains any and all relevant information.
expected_output: >
A fully fledge reports with the mains topics, each with a full section of information.
Formatted as markdown without '```'
agent: reporting_analyst

View File

@@ -0,0 +1,40 @@
from crewai import Agent, Crew, Process, Task
from crewai.project import CrewBase, agent, crew, task
from pydantic import BaseModel
# Uncomment the following line to use an example of a custom tool
# from demo_pipeline.tools.custom_tool import MyCustomTool
# Check our tools documentations for more information on how to use them
# from crewai_tools import SerperDevTool
class UrgencyScore(BaseModel):
urgency_score: int
@CrewBase
class ClassifierCrew:
"""Email Classifier Crew"""
agents_config = "config/agents.yaml"
tasks_config = "config/tasks.yaml"
@agent
def classifier(self) -> Agent:
return Agent(config=self.agents_config["classifier"], verbose=True)
@task
def urgent_task(self) -> Task:
return Task(
config=self.tasks_config["classify_email"],
output_pydantic=UrgencyScore,
)
@crew
def crew(self) -> Crew:
"""Creates the Email Classifier Crew"""
return Crew(
agents=self.agents, # Automatically created by the @agent decorator
tasks=self.tasks, # Automatically created by the @task decorator
process=Process.sequential,
verbose=True,
)

View File

@@ -0,0 +1,7 @@
classifier:
role: >
Email Classifier
goal: >
Classify the email: {email} as urgent or normal from a score of 1 to 10, where 1 is not urgent and 10 is urgent. Return the urgency score only.`
backstory: >
You are a highly efficient and experienced email classifier, trained to quickly assess and classify emails. Your ability to remain calm under pressure and provide concise, actionable responses has made you an invaluable asset in managing normal situations and maintaining smooth operations.

View File

@@ -0,0 +1,7 @@
classify_email:
description: >
Classify the email: {email}
as urgent or normal.
expected_output: >
Classify the email from a scale of 1 to 10, where 1 is not urgent and 10 is urgent. Return the urgency score only.
agent: classifier

View File

@@ -0,0 +1,7 @@
normal_handler:
role: >
Normal Email Processor
goal: >
Process normal emails and create an email to respond to the sender.
backstory: >
You are a highly efficient and experienced normal email handler, trained to quickly assess and respond to normal communications. Your ability to remain calm under pressure and provide concise, actionable responses has made you an invaluable asset in managing normal situations and maintaining smooth operations.

View File

@@ -0,0 +1,6 @@
normal_task:
description: >
Process and respond to normal email quickly.
expected_output: >
An email response to the normal email.
agent: normal_handler

View File

@@ -0,0 +1,36 @@
from crewai import Agent, Crew, Process, Task
from crewai.project import CrewBase, agent, crew, task
# Uncomment the following line to use an example of a custom tool
# from demo_pipeline.tools.custom_tool import MyCustomTool
# Check our tools documentations for more information on how to use them
# from crewai_tools import SerperDevTool
@CrewBase
class NormalCrew:
"""Normal Email Crew"""
agents_config = "config/agents.yaml"
tasks_config = "config/tasks.yaml"
@agent
def normal_handler(self) -> Agent:
return Agent(config=self.agents_config["normal_handler"], verbose=True)
@task
def urgent_task(self) -> Task:
return Task(
config=self.tasks_config["normal_task"],
)
@crew
def crew(self) -> Crew:
"""Creates the Normal Email Crew"""
return Crew(
agents=self.agents, # Automatically created by the @agent decorator
tasks=self.tasks, # Automatically created by the @task decorator
process=Process.sequential,
verbose=True,
)

View File

@@ -0,0 +1,7 @@
urgent_handler:
role: >
Urgent Email Processor
goal: >
Process urgent emails and create an email to respond to the sender.
backstory: >
You are a highly efficient and experienced urgent email handler, trained to quickly assess and respond to time-sensitive communications. Your ability to remain calm under pressure and provide concise, actionable responses has made you an invaluable asset in managing critical situations and maintaining smooth operations.

View File

@@ -0,0 +1,6 @@
urgent_task:
description: >
Process and respond to urgent email quickly.
expected_output: >
An email response to the urgent email.
agent: urgent_handler

View File

@@ -0,0 +1,36 @@
from crewai import Agent, Crew, Process, Task
from crewai.project import CrewBase, agent, crew, task
# Uncomment the following line to use an example of a custom tool
# from demo_pipeline.tools.custom_tool import MyCustomTool
# Check our tools documentations for more information on how to use them
# from crewai_tools import SerperDevTool
@CrewBase
class UrgentCrew:
"""Urgent Email Crew"""
agents_config = "config/agents.yaml"
tasks_config = "config/tasks.yaml"
@agent
def urgent_handler(self) -> Agent:
return Agent(config=self.agents_config["urgent_handler"], verbose=True)
@task
def urgent_task(self) -> Task:
return Task(
config=self.tasks_config["urgent_task"],
)
@crew
def crew(self) -> Crew:
"""Creates the Urgent Email Crew"""
return Crew(
agents=self.agents, # Automatically created by the @agent decorator
tasks=self.tasks, # Automatically created by the @task decorator
process=Process.sequential,
verbose=True,
)

View File

@@ -0,0 +1,75 @@
#!/usr/bin/env python
import asyncio
from crewai.routers.router import Route
from crewai.routers.router import Router
from {{folder_name}}.pipelines.pipeline_classifier import EmailClassifierPipeline
from {{folder_name}}.pipelines.pipeline_normal import NormalPipeline
from {{folder_name}}.pipelines.pipeline_urgent import UrgentPipeline
async def run():
"""
Run the pipeline.
"""
inputs = [
{
"email": """
Subject: URGENT: Marketing Campaign Launch - Immediate Action Required
Dear Team,
I'm reaching out regarding our upcoming marketing campaign that requires your immediate attention and swift action. We're facing a critical deadline, and our success hinges on our ability to mobilize quickly.
Key points:
Campaign launch: 48 hours from now
Target audience: 250,000 potential customers
Expected ROI: 35% increase in Q3 sales
What we need from you NOW:
Final approval on creative assets (due in 3 hours)
Confirmation of media placements (due by end of day)
Last-minute budget allocation for paid social media push
Our competitors are poised to launch similar campaigns, and we must act fast to maintain our market advantage. Delays could result in significant lost opportunities and potential revenue.
Please prioritize this campaign above all other tasks. I'll be available for the next 24 hours to address any concerns or roadblocks.
Let's make this happen!
[Your Name]
Marketing Director
P.S. I'll be scheduling an emergency team meeting in 1 hour to discuss our action plan. Attendance is mandatory.
"""
}
]
pipeline_classifier = EmailClassifierPipeline().create_pipeline()
pipeline_urgent = UrgentPipeline().create_pipeline()
pipeline_normal = NormalPipeline().create_pipeline()
router = Router(
routes={
"high_urgency": Route(
condition=lambda x: x.get("urgency_score", 0) > 7,
pipeline=pipeline_urgent
),
"low_urgency": Route(
condition=lambda x: x.get("urgency_score", 0) <= 7,
pipeline=pipeline_normal
)
},
default=pipeline_normal
)
pipeline = pipeline_classifier >> router
results = await pipeline.kickoff(inputs)
# Process and print results
for result in results:
print(f"Raw output: {result.raw}")
if result.json_dict:
print(f"JSON output: {result.json_dict}")
print("\n")
def main():
asyncio.run(run())
if __name__ == "__main__":
main()

View File

@@ -0,0 +1,24 @@
from crewai import Pipeline
from crewai.project import PipelineBase
from ..crews.classifier_crew.classifier_crew import ClassifierCrew
@PipelineBase
class EmailClassifierPipeline:
def __init__(self):
# Initialize crews
self.classifier_crew = ClassifierCrew().crew()
def create_pipeline(self):
return Pipeline(
stages=[
self.classifier_crew
]
)
async def kickoff(self, inputs):
pipeline = self.create_pipeline()
results = await pipeline.kickoff(inputs)
return results

View File

@@ -0,0 +1,24 @@
from crewai import Pipeline
from crewai.project import PipelineBase
from ..crews.normal_crew.normal_crew import NormalCrew
@PipelineBase
class NormalPipeline:
def __init__(self):
# Initialize crews
self.normal_crew = NormalCrew().crew()
def create_pipeline(self):
return Pipeline(
stages=[
self.normal_crew
]
)
async def kickoff(self, inputs):
pipeline = self.create_pipeline()
results = await pipeline.kickoff(inputs)
return results

View File

@@ -0,0 +1,23 @@
from crewai import Pipeline
from crewai.project import PipelineBase
from ..crews.urgent_crew.urgent_crew import UrgentCrew
@PipelineBase
class UrgentPipeline:
def __init__(self):
# Initialize crews
self.urgent_crew = UrgentCrew().crew()
def create_pipeline(self):
return Pipeline(
stages=[
self.urgent_crew
]
)
async def kickoff(self, inputs):
pipeline = self.create_pipeline()
results = await pipeline.kickoff(inputs)
return results

View File

@@ -0,0 +1,19 @@
[tool.poetry]
name = "{{folder_name}}"
version = "0.1.0"
description = "{{name}} using crewAI"
authors = ["Your Name <you@example.com>"]
[tool.poetry.dependencies]
python = ">=3.10,<=3.13"
crewai = { extras = ["tools"], version = "^0.46.0" }
[tool.poetry.scripts]
{{folder_name}} = "{{folder_name}}.main:main"
train = "{{folder_name}}.main:train"
replay = "{{folder_name}}.main:replay"
test = "{{folder_name}}.main:test"
[build-system]
requires = ["poetry-core"]
build-backend = "poetry.core.masonry.api"

View File

@@ -0,0 +1,12 @@
from crewai_tools import BaseTool
class MyCustomTool(BaseTool):
name: str = "Name of my tool"
description: str = (
"Clear description for what this tool is useful for, you agent will need this information to use it."
)
def _run(self, argument: str) -> str:
# Implementation goes here
return "this is an example of a tool output, ignore it and move along."

18
src/crewai/cli/utils.py Normal file
View File

@@ -0,0 +1,18 @@
import click
def copy_template(src, dst, name, class_name, folder_name):
"""Copy a file from src to dst."""
with open(src, "r") as file:
content = file.read()
# Interpolate the content
content = content.replace("{{name}}", name)
content = content.replace("{{crew_name}}", class_name)
content = content.replace("{{folder_name}}", folder_name)
# Write the interpolated content to the new file
with open(dst, "w") as file:
file.write(content)
click.secho(f" - Created {dst}", fg="green")

View File

@@ -3,7 +3,7 @@ import json
import uuid
from concurrent.futures import Future
from hashlib import md5
from typing import Any, Dict, List, Optional, Tuple, Union
from typing import TYPE_CHECKING, Any, Dict, List, Optional, Tuple, Union
from langchain_core.callbacks import BaseCallbackHandler
from pydantic import (
@@ -32,11 +32,9 @@ from crewai.tasks.conditional_task import ConditionalTask
from crewai.tasks.task_output import TaskOutput
from crewai.telemetry import Telemetry
from crewai.tools.agent_tools import AgentTools
from crewai.types.usage_metrics import UsageMetrics
from crewai.utilities import I18N, FileHandler, Logger, RPMController
from crewai.utilities.constants import (
TRAINED_AGENTS_DATA_FILE,
TRAINING_DATA_FILE,
)
from crewai.utilities.constants import TRAINED_AGENTS_DATA_FILE, TRAINING_DATA_FILE
from crewai.utilities.evaluators.crew_evaluator_handler import CrewEvaluator
from crewai.utilities.evaluators.task_evaluator import TaskEvaluator
from crewai.utilities.formatter import (
@@ -52,6 +50,9 @@ try:
except ImportError:
agentops = None
if TYPE_CHECKING:
from crewai.pipeline.pipeline import Pipeline
class Crew(BaseModel):
"""
@@ -97,6 +98,7 @@ class Crew(BaseModel):
default_factory=TaskOutputStorageHandler
)
name: Optional[str] = Field(default=None)
cache: bool = Field(default=True)
model_config = ConfigDict(arbitrary_types_allowed=True)
tasks: List[Task] = Field(default_factory=list)
@@ -111,7 +113,7 @@ class Crew(BaseModel):
default={"provider": "openai"},
description="Configuration for the embedder to be used for the crew.",
)
usage_metrics: Optional[dict] = Field(
usage_metrics: Optional[UsageMetrics] = Field(
default=None,
description="Metrics for the LLM usage during all tasks execution.",
)
@@ -147,8 +149,8 @@ class Crew(BaseModel):
default=None,
description="Path to the prompt json file to be used for the crew.",
)
output_log_file: Optional[Union[bool, str]] = Field(
default=False,
output_log_file: Optional[str] = Field(
default=None,
description="output_log_file",
)
planning: Optional[bool] = Field(
@@ -453,7 +455,7 @@ class Crew(BaseModel):
if self.planning:
self._handle_crew_planning()
metrics = []
metrics: List[UsageMetrics] = []
if self.process == Process.sequential:
result = self._run_sequential_process()
@@ -463,11 +465,12 @@ class Crew(BaseModel):
raise NotImplementedError(
f"The process '{self.process}' is not implemented yet."
)
metrics += [agent._token_process.get_summary() for agent in self.agents]
self.usage_metrics = {
key: sum([m[key] for m in metrics if m is not None]) for key in metrics[0]
}
self.usage_metrics = UsageMetrics()
for metric in metrics:
self.usage_metrics.add_usage_metrics(metric)
return result
@@ -476,12 +479,7 @@ class Crew(BaseModel):
results: List[CrewOutput] = []
# Initialize the parent crew's usage metrics
total_usage_metrics = {
"total_tokens": 0,
"prompt_tokens": 0,
"completion_tokens": 0,
"successful_requests": 0,
}
total_usage_metrics = UsageMetrics()
for input_data in inputs:
crew = self.copy()
@@ -489,8 +487,7 @@ class Crew(BaseModel):
output = crew.kickoff(inputs=input_data)
if crew.usage_metrics:
for key in total_usage_metrics:
total_usage_metrics[key] += crew.usage_metrics.get(key, 0)
total_usage_metrics.add_usage_metrics(crew.usage_metrics)
results.append(output)
@@ -519,29 +516,10 @@ class Crew(BaseModel):
results = await asyncio.gather(*tasks)
total_usage_metrics = {
"total_tokens": 0,
"prompt_tokens": 0,
"completion_tokens": 0,
"successful_requests": 0,
}
total_usage_metrics = UsageMetrics()
for crew in crew_copies:
if crew.usage_metrics:
for key in total_usage_metrics:
total_usage_metrics[key] += crew.usage_metrics.get(key, 0)
self.usage_metrics = total_usage_metrics
total_usage_metrics = {
"total_tokens": 0,
"prompt_tokens": 0,
"completion_tokens": 0,
"successful_requests": 0,
}
for crew in crew_copies:
if crew.usage_metrics:
for key in total_usage_metrics:
total_usage_metrics[key] += crew.usage_metrics.get(key, 0)
total_usage_metrics.add_usage_metrics(crew.usage_metrics)
self.usage_metrics = total_usage_metrics
self._task_output_handler.reset()
@@ -932,25 +910,18 @@ class Crew(BaseModel):
)
self._telemetry.end_crew(self, final_string_output)
def calculate_usage_metrics(self) -> Dict[str, int]:
def calculate_usage_metrics(self) -> UsageMetrics:
"""Calculates and returns the usage metrics."""
total_usage_metrics = {
"total_tokens": 0,
"prompt_tokens": 0,
"completion_tokens": 0,
"successful_requests": 0,
}
total_usage_metrics = UsageMetrics()
for agent in self.agents:
if hasattr(agent, "_token_process"):
token_sum = agent._token_process.get_summary()
for key in total_usage_metrics:
total_usage_metrics[key] += token_sum.get(key, 0)
total_usage_metrics.add_usage_metrics(token_sum)
if self.manager_agent and hasattr(self.manager_agent, "_token_process"):
token_sum = self.manager_agent._token_process.get_summary()
for key in total_usage_metrics:
total_usage_metrics[key] += token_sum.get(key, 0)
total_usage_metrics.add_usage_metrics(token_sum)
return total_usage_metrics
@@ -969,5 +940,17 @@ class Crew(BaseModel):
evaluator.print_crew_evaluation_result()
def __rshift__(self, other: "Crew") -> "Pipeline":
"""
Implements the >> operator to add another Crew to an existing Pipeline.
"""
from crewai.pipeline.pipeline import Pipeline
if not isinstance(other, Crew):
raise TypeError(
f"Unsupported operand type for >>: '{type(self).__name__}' and '{type(other).__name__}'"
)
return Pipeline(stages=[self, other])
def __repr__(self):
return f"Crew(id={self.id}, process={self.process}, number_of_agents={len(self.agents)}, number_of_tasks={len(self.tasks)})"

View File

@@ -5,6 +5,7 @@ from pydantic import BaseModel, Field
from crewai.tasks.output_format import OutputFormat
from crewai.tasks.task_output import TaskOutput
from crewai.types.usage_metrics import UsageMetrics
class CrewOutput(BaseModel):
@@ -20,9 +21,7 @@ class CrewOutput(BaseModel):
tasks_output: list[TaskOutput] = Field(
description="Output of each task", default=[]
)
token_usage: Dict[str, Any] = Field(
description="Processed token summary", default={}
)
token_usage: UsageMetrics = Field(description="Processed token summary", default={})
@property
def json(self) -> Optional[str]:

View File

@@ -5,12 +5,12 @@ import os
import shutil
from typing import Any, Dict, List, Optional
from embedchain import App
from embedchain.llm.base import BaseLlm
from embedchain.vectordb.chroma import InvalidDimensionException
from crewai.memory.storage.interface import Storage
from crewai.utilities.paths import db_storage_path
from embedchain import App
from embedchain.llm.base import BaseLlm
from embedchain.models.data_type import DataType
from embedchain.vectordb.chroma import InvalidDimensionException
@contextlib.contextmanager
@@ -101,8 +101,7 @@ class RAGStorage(Storage):
return [r for r in results if r["metadata"]["score"] >= score_threshold]
def _generate_embedding(self, text: str, metadata: Dict[str, Any]) -> Any:
with suppress_logging():
self.app.add(text, data_type="text", metadata=metadata)
self.app.add(text, data_type=DataType.TEXT, metadata=metadata)
def reset(self) -> None:
try:

View File

@@ -0,0 +1,3 @@
from crewai.pipeline.pipeline import Pipeline
from crewai.pipeline.pipeline_kickoff_result import PipelineKickoffResult
from crewai.pipeline.pipeline_output import PipelineOutput

View File

@@ -0,0 +1,405 @@
import asyncio
import copy
from typing import Any, Dict, List, Tuple, Union
from pydantic import BaseModel, Field, model_validator
from crewai.crew import Crew
from crewai.crews.crew_output import CrewOutput
from crewai.pipeline.pipeline_kickoff_result import PipelineKickoffResult
from crewai.routers.router import Router
from crewai.types.usage_metrics import UsageMetrics
Trace = Union[Union[str, Dict[str, Any]], List[Union[str, Dict[str, Any]]]]
PipelineStage = Union[Crew, List[Crew], Router]
"""
Developer Notes:
This module defines a Pipeline class that represents a sequence of operations (stages)
to process inputs. Each stage can be either sequential or parallel, and the pipeline
can process multiple kickoffs concurrently.
Core Loop Explanation:
1. The `process_kickoffs` method processes multiple kickoffs in parallel, each going through
all pipeline stages.
2. The `process_single_kickoff` method handles the processing of a single kickouff through
all stages, updating metrics and input data along the way.
3. The `_process_stage` method determines whether a stage is sequential or parallel
and processes it accordingly.
4. The `_process_single_crew` and `_process_parallel_crews` methods handle the
execution of single and parallel crew stages.
5. The `_update_metrics_and_input` method updates usage metrics and the current input
with the outputs from a stage.
6. The `_build_pipeline_kickoff_results` method constructs the final results of the
pipeline kickoff, including traces and outputs.
Handling Traces and Crew Outputs:
- During the processing of stages, we handle the results (traces and crew outputs)
for all stages except the last one differently from the final stage.
- For intermediate stages, the primary focus is on passing the input data between stages.
This involves merging the output dictionaries from all crews in a stage into a single
dictionary and passing it to the next stage. This merged dictionary allows for smooth
data flow between stages.
- For the final stage, in addition to passing the input data, we also need to prepare
the final outputs and traces to be returned as the overall result of the pipeline kickoff.
In this case, we do not merge the results, as each result needs to be included
separately in its own pipeline kickoff result.
Pipeline Terminology:
- Pipeline: The overall structure that defines a sequence of operations.
- Stage: A distinct part of the pipeline, which can be either sequential or parallel.
- Kickoff: A specific execution of the pipeline for a given set of inputs, representing a single instance of processing through the pipeline.
- Branch: Parallel executions within a stage (e.g., concurrent crew operations).
- Trace: The journey of an individual input through the entire pipeline.
Example pipeline structure:
crew1 >> crew2 >> crew3
This represents a pipeline with three sequential stages:
1. crew1 is the first stage, which processes the input and passes its output to crew2.
2. crew2 is the second stage, which takes the output from crew1 as its input, processes it, and passes its output to crew3.
3. crew3 is the final stage, which takes the output from crew2 as its input and produces the final output of the pipeline.
Each input creates its own kickoff, flowing through all stages of the pipeline.
Multiple kickoffss can be processed concurrently, each following the defined pipeline structure.
Another example pipeline structure:
crew1 >> [crew2, crew3] >> crew4
This represents a pipeline with three stages:
1. A sequential stage (crew1)
2. A parallel stage with two branches (crew2 and crew3 executing concurrently)
3. Another sequential stage (crew4)
Each input creates its own kickoff, flowing through all stages of the pipeline.
Multiple kickoffs can be processed concurrently, each following the defined pipeline structure.
"""
class Pipeline(BaseModel):
stages: List[PipelineStage] = Field(
..., description="List of crews representing stages to be executed in sequence"
)
@model_validator(mode="before")
@classmethod
def validate_stages(cls, values):
stages = values.get("stages", [])
def check_nesting_and_type(item, depth=0):
if depth > 1:
raise ValueError("Double nesting is not allowed in pipeline stages")
if isinstance(item, list):
for sub_item in item:
check_nesting_and_type(sub_item, depth + 1)
elif not isinstance(item, (Crew, Router)):
raise ValueError(
f"Expected Crew instance, Router instance, or list of Crews, got {type(item)}"
)
for stage in stages:
check_nesting_and_type(stage)
return values
async def kickoff(
self, inputs: List[Dict[str, Any]]
) -> List[PipelineKickoffResult]:
"""
Processes multiple runs in parallel, each going through all pipeline stages.
Args:
inputs (List[Dict[str, Any]]): List of inputs for each run.
Returns:
List[PipelineKickoffResult]: List of results from each run.
"""
pipeline_results: List[PipelineKickoffResult] = []
# Process all runs in parallel
all_run_results = await asyncio.gather(
*(self.process_single_kickoff(input_data) for input_data in inputs)
)
# Flatten the list of lists into a single list of results
pipeline_results.extend(
result for run_result in all_run_results for result in run_result
)
return pipeline_results
async def process_single_kickoff(
self, kickoff_input: Dict[str, Any]
) -> List[PipelineKickoffResult]:
"""
Processes a single run through all pipeline stages.
Args:
input (Dict[str, Any]): The input for the run.
Returns:
List[PipelineKickoffResult]: The results of processing the run.
"""
initial_input = copy.deepcopy(kickoff_input)
current_input = copy.deepcopy(kickoff_input)
stages = self._copy_stages()
pipeline_usage_metrics: Dict[str, UsageMetrics] = {}
all_stage_outputs: List[List[CrewOutput]] = []
traces: List[List[Union[str, Dict[str, Any]]]] = [[initial_input]]
stage_index = 0
while stage_index < len(stages):
stage = stages[stage_index]
stage_input = copy.deepcopy(current_input)
if isinstance(stage, Router):
next_pipeline, route_taken = stage.route(stage_input)
stages = (
stages[: stage_index + 1]
+ list(next_pipeline.stages)
+ stages[stage_index + 1 :]
)
traces.append([{"route_taken": route_taken}])
stage_index += 1
continue
stage_outputs, stage_trace = await self._process_stage(stage, stage_input)
self._update_metrics_and_input(
pipeline_usage_metrics, current_input, stage, stage_outputs
)
traces.append(stage_trace)
all_stage_outputs.append(stage_outputs)
stage_index += 1
return self._build_pipeline_kickoff_results(
all_stage_outputs, traces, pipeline_usage_metrics
)
async def _process_stage(
self, stage: PipelineStage, current_input: Dict[str, Any]
) -> Tuple[List[CrewOutput], List[Union[str, Dict[str, Any]]]]:
"""
Processes a single stage of the pipeline, which can be either sequential or parallel.
Args:
stage (Union[Crew, List[Crew]]): The stage to process.
current_input (Dict[str, Any]): The input for the stage.
Returns:
Tuple[List[CrewOutput], List[Union[str, Dict[str, Any]]]]: The outputs and trace of the stage.
"""
if isinstance(stage, Crew):
return await self._process_single_crew(stage, current_input)
elif isinstance(stage, list) and all(isinstance(crew, Crew) for crew in stage):
return await self._process_parallel_crews(stage, current_input)
else:
raise ValueError(f"Unsupported stage type: {type(stage)}")
async def _process_single_crew(
self, crew: Crew, current_input: Dict[str, Any]
) -> Tuple[List[CrewOutput], List[Union[str, Dict[str, Any]]]]:
"""
Processes a single crew.
Args:
crew (Crew): The crew to process.
current_input (Dict[str, Any]): The input for the crew.
Returns:
Tuple[List[CrewOutput], List[Union[str, Dict[str, Any]]]]: The output and trace of the crew.
"""
output = await crew.kickoff_async(inputs=current_input)
return [output], [crew.name or str(crew.id)]
async def _process_parallel_crews(
self, crews: List[Crew], current_input: Dict[str, Any]
) -> Tuple[List[CrewOutput], List[Union[str, Dict[str, Any]]]]:
"""
Processes multiple crews in parallel.
Args:
crews (List[Crew]): The list of crews to process in parallel.
current_input (Dict[str, Any]): The input for the crews.
Returns:
Tuple[List[CrewOutput], List[Union[str, Dict[str, Any]]]]: The outputs and traces of the crews.
"""
parallel_outputs = await asyncio.gather(
*[crew.kickoff_async(inputs=current_input) for crew in crews]
)
return parallel_outputs, [crew.name or str(crew.id) for crew in crews]
def _update_metrics_and_input(
self,
usage_metrics: Dict[str, UsageMetrics],
current_input: Dict[str, Any],
stage: PipelineStage,
outputs: List[CrewOutput],
) -> None:
"""
Updates metrics and current input with the outputs of a stage.
Args:
usage_metrics (Dict[str, Any]): The usage metrics to update.
current_input (Dict[str, Any]): The current input to update.
stage (Union[Crew, List[Crew]]): The stage that was processed.
outputs (List[CrewOutput]): The outputs of the stage.
"""
if isinstance(stage, Crew):
usage_metrics[stage.name or str(stage.id)] = outputs[0].token_usage
current_input.update(outputs[0].to_dict())
elif isinstance(stage, list) and all(isinstance(crew, Crew) for crew in stage):
for crew, output in zip(stage, outputs):
usage_metrics[crew.name or str(crew.id)] = output.token_usage
current_input.update(output.to_dict())
else:
raise ValueError(f"Unsupported stage type: {type(stage)}")
def _build_pipeline_kickoff_results(
self,
all_stage_outputs: List[List[CrewOutput]],
traces: List[List[Union[str, Dict[str, Any]]]],
token_usage: Dict[str, UsageMetrics],
) -> List[PipelineKickoffResult]:
"""
Builds the results of a pipeline run.
Args:
all_stage_outputs (List[List[CrewOutput]]): All stage outputs.
traces (List[List[Union[str, Dict[str, Any]]]]): All traces.
token_usage (Dict[str, Any]): Token usage metrics.
Returns:
List[PipelineKickoffResult]: The results of the pipeline run.
"""
formatted_traces = self._format_traces(traces)
formatted_crew_outputs = self._format_crew_outputs(all_stage_outputs)
return [
PipelineKickoffResult(
token_usage=token_usage,
trace=formatted_trace,
raw=crews_outputs[-1].raw,
pydantic=crews_outputs[-1].pydantic,
json_dict=crews_outputs[-1].json_dict,
crews_outputs=crews_outputs,
)
for crews_outputs, formatted_trace in zip(
formatted_crew_outputs, formatted_traces
)
]
def _format_traces(
self, traces: List[List[Union[str, Dict[str, Any]]]]
) -> List[List[Trace]]:
"""
Formats the traces of a pipeline run.
Args:
traces (List[List[Union[str, Dict[str, Any]]]]): The traces to format.
Returns:
List[List[Trace]]: The formatted traces.
"""
formatted_traces: List[Trace] = self._format_single_trace(traces[:-1])
return self._format_multiple_traces(formatted_traces, traces[-1])
def _format_single_trace(
self, traces: List[List[Union[str, Dict[str, Any]]]]
) -> List[Trace]:
"""
Formats single traces.
Args:
traces (List[List[Union[str, Dict[str, Any]]]]): The traces to format.
Returns:
List[Trace]: The formatted single traces.
"""
formatted_traces: List[Trace] = []
for trace in traces:
formatted_traces.append(trace[0] if len(trace) == 1 else trace)
return formatted_traces
def _format_multiple_traces(
self,
formatted_traces: List[Trace],
final_trace: List[Union[str, Dict[str, Any]]],
) -> List[List[Trace]]:
"""
Formats multiple traces.
Args:
formatted_traces (List[Trace]): The formatted single traces.
final_trace (List[Union[str, Dict[str, Any]]]): The final trace to format.
Returns:
List[List[Trace]]: The formatted multiple traces.
"""
traces_to_return: List[List[Trace]] = []
if len(final_trace) == 1:
formatted_traces.append(final_trace[0])
traces_to_return.append(formatted_traces)
else:
for trace in final_trace:
copied_traces = formatted_traces.copy()
copied_traces.append(trace)
traces_to_return.append(copied_traces)
return traces_to_return
def _format_crew_outputs(
self, all_stage_outputs: List[List[CrewOutput]]
) -> List[List[CrewOutput]]:
"""
Formats the outputs of all stages into a list of crew outputs.
Args:
all_stage_outputs (List[List[CrewOutput]]): All stage outputs.
Returns:
List[List[CrewOutput]]: Formatted crew outputs.
"""
crew_outputs: List[CrewOutput] = [
output
for stage_outputs in all_stage_outputs[:-1]
for output in stage_outputs
]
return [crew_outputs + [output] for output in all_stage_outputs[-1]]
def _copy_stages(self):
"""Create a deep copy of the Pipeline's stages."""
new_stages = []
for stage in self.stages:
if isinstance(stage, list):
new_stages.append(
[
crew.copy() if hasattr(crew, "copy") else copy.deepcopy(crew)
for crew in stage
]
)
elif hasattr(stage, "copy"):
new_stages.append(stage.copy())
else:
new_stages.append(copy.deepcopy(stage))
return new_stages
def __rshift__(self, other: PipelineStage) -> "Pipeline":
"""
Implements the >> operator to add another Stage (Crew or List[Crew]) to an existing Pipeline.
Args:
other (Any): The stage to add.
Returns:
Pipeline: A new pipeline with the added stage.
"""
if isinstance(other, (Crew, Router)) or (
isinstance(other, list) and all(isinstance(item, Crew) for item in other)
):
return type(self)(stages=self.stages + [other])
else:
raise TypeError(
f"Unsupported operand type for >>: '{type(self).__name__}' and '{type(other).__name__}'"
)

View File

@@ -0,0 +1,61 @@
import json
import uuid
from typing import Any, Dict, List, Optional, Union
from pydantic import UUID4, BaseModel, Field
from crewai.crews.crew_output import CrewOutput
from crewai.types.usage_metrics import UsageMetrics
class PipelineKickoffResult(BaseModel):
"""Class that represents the result of a pipeline run."""
id: UUID4 = Field(
default_factory=uuid.uuid4,
frozen=True,
description="Unique identifier for the object, not set by user.",
)
raw: str = Field(description="Raw output of the pipeline run", default="")
pydantic: Any = Field(
description="Pydantic output of the pipeline run", default=None
)
json_dict: Union[Dict[str, Any], None] = Field(
description="JSON dict output of the pipeline run", default={}
)
token_usage: Dict[str, UsageMetrics] = Field(
description="Token usage for each crew in the run"
)
trace: List[Any] = Field(
description="Trace of the journey of inputs through the run"
)
crews_outputs: List[CrewOutput] = Field(
description="Output from each crew in the run",
default=[],
)
@property
def json(self) -> Optional[str]:
if self.crews_outputs[-1].tasks_output[-1].output_format != "json":
raise ValueError(
"No JSON output found in the final task of the final crew. Please make sure to set the output_json property in the final task in your crew."
)
return json.dumps(self.json_dict)
def to_dict(self) -> Dict[str, Any]:
"""Convert json_output and pydantic_output to a dictionary."""
output_dict = {}
if self.json_dict:
output_dict.update(self.json_dict)
elif self.pydantic:
output_dict.update(self.pydantic.model_dump())
return output_dict
def __str__(self):
if self.pydantic:
return str(self.pydantic)
if self.json_dict:
return str(self.json_dict)
return self.raw

View File

@@ -0,0 +1,20 @@
import uuid
from typing import List
from pydantic import UUID4, BaseModel, Field
from crewai.pipeline.pipeline_kickoff_result import PipelineKickoffResult
class PipelineOutput(BaseModel):
id: UUID4 = Field(
default_factory=uuid.uuid4,
frozen=True,
description="Unique identifier for the object, not set by user.",
)
run_results: List[PipelineKickoffResult] = Field(
description="List of results for each run through the pipeline", default=[]
)
def add_run_result(self, result: PipelineKickoffResult):
self.run_results.append(result)

View File

@@ -1,15 +1,17 @@
from .annotations import (
agent,
cache_handler,
callback,
crew,
task,
llm,
output_json,
output_pydantic,
pipeline,
task,
tool,
callback,
llm,
cache_handler,
)
from .crew_base import CrewBase
from .pipeline_base import PipelineBase
__all__ = [
"agent",
@@ -20,6 +22,8 @@ __all__ = [
"tool",
"callback",
"CrewBase",
"PipelineBase",
"llm",
"cache_handler",
"pipeline",
]

View File

@@ -1,14 +1,4 @@
def memoize(func):
cache = {}
def memoized_func(*args, **kwargs):
key = (args, tuple(kwargs.items()))
if key not in cache:
cache[key] = func(*args, **kwargs)
return cache[key]
memoized_func.__dict__.update(func.__dict__)
return memoized_func
from crewai.project.utils import memoize
def task(func):
@@ -61,6 +51,21 @@ def cache_handler(func):
return memoize(func)
def stage(func):
func.is_stage = True
return memoize(func)
def router(func):
func.is_router = True
return memoize(func)
def pipeline(func):
func.is_pipeline = True
return memoize(func)
def crew(func):
def wrapper(self, *args, **kwargs):
instantiated_tasks = []

View File

@@ -24,6 +24,7 @@ def CrewBase(cls):
original_agents_config_path = getattr(
cls, "agents_config", "config/agents.yaml"
)
original_tasks_config_path = getattr(cls, "tasks_config", "config/tasks.yaml")
def __init__(self, *args, **kwargs):
@@ -37,9 +38,11 @@ def CrewBase(cls):
self.agents_config = self.load_yaml(
os.path.join(self.base_directory, self.original_agents_config_path)
)
self.tasks_config = self.load_yaml(
os.path.join(self.base_directory, self.original_tasks_config_path)
)
self.map_all_agent_variables()
self.map_all_task_variables()

View File

@@ -0,0 +1,58 @@
from typing import Callable, Dict
from pydantic import ConfigDict
from crewai.crew import Crew
from crewai.pipeline.pipeline import Pipeline
from crewai.routers.router import Router
# TODO: Could potentially remove. Need to check with @joao and @gui if this is needed for CrewAI+
def PipelineBase(cls):
class WrappedClass(cls):
model_config = ConfigDict(arbitrary_types_allowed=True)
is_pipeline_class: bool = True
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
self.stages = []
self._map_pipeline_components()
def _get_all_functions(self):
return {
name: getattr(self, name)
for name in dir(self)
if callable(getattr(self, name))
}
def _filter_functions(
self, functions: Dict[str, Callable], attribute: str
) -> Dict[str, Callable]:
return {
name: func
for name, func in functions.items()
if hasattr(func, attribute)
}
def _map_pipeline_components(self):
all_functions = self._get_all_functions()
crew_functions = self._filter_functions(all_functions, "is_crew")
router_functions = self._filter_functions(all_functions, "is_router")
for stage_attr in dir(self):
stage = getattr(self, stage_attr)
if isinstance(stage, (Crew, Router)):
self.stages.append(stage)
elif callable(stage) and hasattr(stage, "is_crew"):
self.stages.append(crew_functions[stage_attr]())
elif callable(stage) and hasattr(stage, "is_router"):
self.stages.append(router_functions[stage_attr]())
elif isinstance(stage, list) and all(
isinstance(item, Crew) for item in stage
):
self.stages.append(stage)
def build_pipeline(self) -> Pipeline:
return Pipeline(stages=self.stages)
return WrappedClass

View File

@@ -0,0 +1,11 @@
def memoize(func):
cache = {}
def memoized_func(*args, **kwargs):
key = (args, tuple(kwargs.items()))
if key not in cache:
cache[key] = func(*args, **kwargs)
return cache[key]
memoized_func.__dict__.update(func.__dict__)
return memoized_func

View File

@@ -0,0 +1 @@
from crewai.routers.router import Router

View File

@@ -0,0 +1,90 @@
from copy import deepcopy
from typing import Any, Callable, Dict, Generic, Tuple, TypeVar
from pydantic import BaseModel, Field, PrivateAttr
T = TypeVar("T", bound=Dict[str, Any])
U = TypeVar("U")
class Route(Generic[T, U]):
condition: Callable[[T], bool]
pipeline: U
def __init__(self, condition: Callable[[T], bool], pipeline: U):
self.condition = condition
self.pipeline = pipeline
class Router(BaseModel, Generic[T, U]):
routes: Dict[str, Route[T, U]] = Field(
default_factory=dict,
description="Dictionary of route names to (condition, pipeline) tuples",
)
default: U = Field(..., description="Default pipeline if no conditions are met")
_route_types: Dict[str, type] = PrivateAttr(default_factory=dict)
model_config = {"arbitrary_types_allowed": True}
def __init__(self, routes: Dict[str, Route[T, U]], default: U, **data):
super().__init__(routes=routes, default=default, **data)
self._check_copyable(default)
for name, route in routes.items():
self._check_copyable(route.pipeline)
self._route_types[name] = type(route.pipeline)
@staticmethod
def _check_copyable(obj):
if not hasattr(obj, "copy") or not callable(getattr(obj, "copy")):
raise ValueError(f"Object of type {type(obj)} must have a 'copy' method")
def add_route(
self,
name: str,
condition: Callable[[T], bool],
pipeline: U,
) -> "Router[T, U]":
"""
Add a named route with its condition and corresponding pipeline to the router.
Args:
name: A unique name for this route
condition: A function that takes a dictionary input and returns a boolean
pipeline: The Pipeline to execute if the condition is met
Returns:
The Router instance for method chaining
"""
self._check_copyable(pipeline)
self.routes[name] = Route(condition=condition, pipeline=pipeline)
self._route_types[name] = type(pipeline)
return self
def route(self, input_data: T) -> Tuple[U, str]:
"""
Evaluate the input against the conditions and return the appropriate pipeline.
Args:
input_data: The input dictionary to be evaluated
Returns:
A tuple containing the next Pipeline to be executed and the name of the route taken
"""
for name, route in self.routes.items():
if route.condition(input_data):
return route.pipeline, name
return self.default, "default"
def copy(self) -> "Router[T, U]":
"""Create a deep copy of the Router."""
new_routes = {
name: Route(
condition=deepcopy(route.condition),
pipeline=route.pipeline.copy(), # type: ignore
)
for name, route in self.routes.items()
}
new_default = self.default.copy() # type: ignore
return Router(routes=new_routes, default=new_default)

View File

@@ -47,6 +47,7 @@ class Task(BaseModel):
tools_errors: int = 0
delegations: int = 0
i18n: I18N = I18N()
name: Optional[str] = Field(default=None)
prompt_context: Optional[str] = None
description: str = Field(description="Description of the actual task.")
expected_output: str = Field(

View File

View File

@@ -0,0 +1,36 @@
from pydantic import BaseModel, Field
class UsageMetrics(BaseModel):
"""
Model to track usage metrics for the crew's execution.
Attributes:
total_tokens: Total number of tokens used.
prompt_tokens: Number of tokens used in prompts.
completion_tokens: Number of tokens used in completions.
successful_requests: Number of successful requests made.
"""
total_tokens: int = Field(default=0, description="Total number of tokens used.")
prompt_tokens: int = Field(
default=0, description="Number of tokens used in prompts."
)
completion_tokens: int = Field(
default=0, description="Number of tokens used in completions."
)
successful_requests: int = Field(
default=0, description="Number of successful requests made."
)
def add_usage_metrics(self, usage_metrics: "UsageMetrics"):
"""
Add the usage metrics from another UsageMetrics object.
Args:
usage_metrics (UsageMetrics): The usage metrics to add.
"""
self.total_tokens += usage_metrics.total_tokens
self.prompt_tokens += usage_metrics.prompt_tokens
self.completion_tokens += usage_metrics.completion_tokens
self.successful_requests += usage_metrics.successful_requests