feature: Separate dialog history of each agent

[jinhyun95]

This PR separate dialog history of multiple agents.

• As-is) Talking to multiple agents using a single user ID leads to mixed dialog history.
• To-be) Service adds agent ID to thread ID to separate dialog histories in _handle_input.
• Related to: Independent Chat Memory for Each Agent with Unique ThreadID #188

[JoshuaC215]

I'm supportive of separating history by agent in a better way than provided today. But I'm not immediately a fan of the other changes. If you want to open an issue for removing default agent and forcing the user to always specify an agent, we could see what others in the community think and get some discussion. Probably should be separate PRs.

[jinhyun95]

Thanks for the review :)
I reverted the changes regarding default agent and APIs and amended the commit (08d79dc)
thread_id is now handled in service, specifically _handle_input.
PR description is edited accordingly.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 62.51%. Comparing base (2e6c622) to head (08d79dc).
⚠️ Report is 21 commits behind head on main.

Files with missing lines	Coverage Δ
src/service/service.py	79.50% <100.00%> (ø)

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[JoshuaC215]

So I think the simplified change creates a few problems, particularly in the interaction with the streamlit app (maybe we need to add tests for those)

1. The /history endpoint just looks naively for the thread_id, so I think it will always return nothing since it doesn't match the format threads are being saved in. To resolve this, we might use the same pattern used by /invoke and /stream where you can optionally insert the {agent_id} in the path. If it's omitted, then it returns the history of the default agent.
2. The streamlit app session_state isn't being updated when I select a new agent, so the previous history messages will stay on the users screen even though they will no longer be visible to the agent. This is likely to confuse the user. I guess to provide persistence, we would need to query the history endpoint each time you change agents in case you're returning to an agent you talked with before.

I guess we would ideally provide some tests for this functionality as well, since the tests are passing even though the user interaction behavior will be broken.

[jinhyun95]

So I think the simplified change creates a few problems, particularly in the interaction with the streamlit app (maybe we need to add tests for those)

1. The `/history` endpoint just looks naively for the thread_id, so I think it will always return nothing since it doesn't match the format threads are being saved in. To resolve this, we might use the same pattern used by `/invoke` and `/stream` where you can optionally insert the `{agent_id}` in the path. If it's omitted, then it returns the history of the default agent.

2. The streamlit app session_state isn't being updated when I select a new agent, so the previous history messages will stay on the users screen even though they will no longer be visible to the agent. This is likely to confuse the user. I guess to provide persistence, we would need to query the history endpoint each time you change agents in case you're returning to an agent you talked with before.

I guess we would ideally provide some tests for this functionality as well, since the tests are passing even though the user interaction behavior will be broken.

While working on (1), I noticed that chatbot agent's state only retains the latest chat message due to the agent's lack of explicit state management.
Would you mind if I replace the agent with that with MessageState?

[JoshuaC215]

Happy to review but it should be a separate PR

[andermirik]

thread_id must be uuid4 by langgraph api spec and langchain agent protocol.

[JoshuaC215]

Stale, feel free to re-open if you pick it back up