CG-10805 & CG-10806: Migrate Everything to Langgraph + Multi-LLM Config (openai + anthropic)

[tawsifkamal]

• migrated everything over to langgarph

• now uses LLM class for everything including the model in semantic_edit.py

• no more AgentExecutor

• The syntax for codeagent stays the same

from codegen import CodeAgent
from codegen import Codebase

codebase = Codebase("./")
agent = CodeAgent(codebase=codebase, thread_id=1)
output = agent.run("Go into one of the files")

• the agent streams each step agent.stream now instead of calling invoke so that intermediate steps can be printed

• If you want to directly invoke the agent, then use thread_id instead + set debug to true for the create_codebase_agent function so that logs get printed. However, streaming is preferred and outputs nicer logs as well. agent.stream does NOT mean streaming each token! Streaming indicates getting each agent action one by one. agent.invoke won't output until the the agent finishes the execution

• langgraph uses threads to keep track of the conversation history

output = agent.invoke({"messages": [("user", "Go into one of the files")]}, config={"configurable": {"thread_id": 1}})

If you want to call invoke for whatever reason, then you can set debug flag to true for printed debug statements.
However, streaming shows nicer outputs. Streaming also does NOT mean streaming each token. agent.stream means to stream each step, and thus we can print each step however we want.

# add the debug flag to true if you want to Debug intermediate steps when calling invoke (this does not apply to agent.stream())
create_react_agent(model=llm, tools=tools, prompt=system_message, checkpointer=memory, debug=debug)

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
1 out of 2 committers have signed the CLA.

✅ tawsifkamal
❌ codegen-bot

---

codegen-bot seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

✅ All tests successful. No failed tests found.

Additional details and impacted files

[kopekC]

The error handling in the LLM class could be improved by adding more specific error types and messages for common failure scenarios (rate limits, token limits, etc). Consider adding a custom exception hierarchy.

[kopekC]

Consider adding validation for model names against a list of supported models for each provider. This would help catch configuration errors early.

[kopekC]

The memory implementation could benefit from configurable cleanup/expiry policies to prevent memory leaks in long-running sessions.

[kopekC]

The docstrings for the new LLM class methods could be more detailed, especially regarding the expected behavior when switching between different model providers.

[kopekC]

While the LLM class has good test coverage, consider adding integration tests that verify the interaction between the LLM, agent, and memory components.

[kopekC]

Overall Assessment:

This PR represents a significant improvement to the codebase's LLM integration architecture. The changes make the system more flexible and maintainable by:

1. Unifying the LLM interface across providers
2. Improving configuration management
3. Adding proper memory/session handling
4. Enhancing test coverage

The code quality is good, with clear separation of concerns and proper error handling. The migration to langgraph's ReAct agent is a positive architectural decision that should improve reliability and maintainability.

Suggested improvements have been noted in inline comments, focusing on:

• Enhanced error handling
• Model validation
• Memory management policies
• Documentation completeness
• Integration testing

These suggestions are relatively minor and could be addressed in follow-up PRs. The current changes are well-structured and ready for merging after addressing any critical inline comments.

[github-actions]

🎉 This PR is included in version 0.32.0 🎉

The release is available on GitHub release

Your semantic-release bot 📦🚀

[PramaYudhistira]

yoooooo lets get it bro 🔥🔥🔥🔥🔥🔥🔥