* fixed bug for manager overriding task agent and then added pydanic valditors to sequential when no agent is added to task
* better test and fixed task.agent logic
* fixed tests and better validator message
* added validator for async_execution true in tasks whenever in hierarchical run
* WIP. Figuring out disconnect issue.
* Cleaned up logs now that I've isolated the issue to the LLM
* more wip.
* WIP. It looks like usage metrics has always been broken for async
* Update parent crew who is managing for_each loop
* Merge in main to bugfix/kickoff-for-each-usage-metrics
* Clean up code for review
* Add new tests
* Final cleanup. Ready for review.
* Moving copy functionality from Agent to BaseAgent
* Fix renaming issue
* Fix linting errors
* use BaseAgent instead of Agent where applicable
* better spacing
* works with llama index
* works on langchain custom just need delegation to work
* cleanup for custom_agent class
* works with different argument expectations for agent_executor
* cleanup for hierarchial process, better agent_executor args handler and added to the crew agent doc page
* removed code examples for langchain + llama index, added to docs instead
* added key output if return is not a str for and added some tests
* added hinting for CustomAgent class
* removed pass as it was not needed
* closer just need to figuire ou agentTools
* running agents - llamaindex and langchain with base agent
* some cleanup on baseAgent
* minimum for agent to run for base class and ensure it works with hierarchical process
* cleanup for original agent to take on BaseAgent class
* Agent takes on langchainagent and cleanup across
* token handling working for usage_metrics to continue working
* installed llama-index, updated docs and added better name
* fixed some type errors
* base agent holds token_process
* heirarchail process uses proper tools and no longer relies on hasattr for token_processes
* removal of test_custom_agent_executions
* this fixes copying agents
* leveraging an executor class for trigger llamaindex agent
* llama index now has ask_human
* executor mixins added
* added output converter base class
* type listed
* cleanup for output conversions and tokenprocess eliminated redundancy
* properly handling tokens
* simplified token calc handling
* original agent with base agent builder structure setup
* better docs
* no more llama-index dep
* cleaner docs
* test fixes
* poetry reverts and better docs
* base_agent_tools set for third party agents
* updated task and test fix
* removed hyphen in co-workers
* Fix issue with AgentTool agent selection. The LLM included double quotes in the agent name which messed up the string comparison. Added additional types. Cleaned up error messaging.
* Remove duplicate import
* Improve explanation
* Revert poetry.lock changes
* Fix missing line in poetry.lock
---------
Co-authored-by: madmag77 <goncharov.artemv@gmail.com>
* mentioning ollama on the docs as embedder
* lowering barrier to match tool with simialr name
* Fixing agent tools to support co_worker
* Adding new tests
* Fixing type"
* updating tests
* fixing conflict
* log_file
feature: added a new parameter for crew that creates a txt file to log agent execution
* unit tests and documentation
unit test if file is created but not what is inside the file
* feature: human input per task
* Update executor.py
* Update executor.py
* Update executor.py
* Update executor.py
* Update executor.py
* feat: change human input for unit testing
added documentation and unit test
* Create test_agent_human_input.yaml
add yaml for test
---------
Co-authored-by: João Moura <joaomdmoura@gmail.com>
This now will allow to add a max_inter option to agents while also making sure to force the agent to give it's best final answer before running out of it's max_inter.
* Refactoring task cache to be a tool
The previous implementation of the task caching system was early exiting
the agent executor due to the fact it was returning an AgentFinish object.
This now refactors it to use a cache specific tool that is dynamically
added and forced into the agent in case of a task execution that was
already executed with the same input.
* Adding tool caching a loop execution prevention.
This adds some guardrails, to both prevent the same tool to be used
consecutively and also caching tool's results across the entire crew
so it cuts down execution time and eventual LLM calls.
This plays a huge role for smaller opensource models that usually fall
into those behaviors patterns.
It also includes some smaller improvements around the tool prompt and
agent tools, all with the same intention of guiding models into
better conform with agent instructions.
