Feature/agentic chat 2935

[Romamo]

I started to implement the agentic chat version.

It's partially completed because the implementation requires changes across multiple files, and it's not ready to be merged yet.

Please review, run and test.

Here it is:

graph_stream = create_agent(
    llm_mini_stream,
    [
        get_memories,
        get_conversations,
    ],
    checkpointer=checkpointer
)

How to run:

1. Checkout the branch
2. Upgrade langchain:

pip install --upgrade langchain==1.0.0a10
pip install --upgrade langchain-core==1.0.0a6

3. Add a new line to .env:

CHAT_AGENTIC=true

Implemented so far:

1. New agentic chat with streaming only(execute_graph_chat_stream)
2. Tools: get_memories and get_conversations only

My plan:

1. Use new langchain agent v1
2. Keep the system design compatible to current implementation(minimum changes)
3. Continue to implement other important tools: save_memory

/claim #2935

[beastoin]

@Romamo what's stopping you? just go ahead and finish this, man!

tip: check routers > integration. it includes the tools for omigpt, and you know it's just simple like this. you've got it, i guess.

[Romamo]

It's almost done:
+get_memories
+get_conversations
+apps(personas)
+citations(sometimes not returned by LLM - requires more work on prompt or data organization)
?difference in token spend
-ask_for_nps(agentic mode is not so straigth)

Persona has no access to memories and conversations - is it ok?

[beastoin]

hi man, can you launch a quick demo and note what works / doesn’t work, along with your further concerns?

i checked your pr, but sorry, i cannot get this information closely, so i need your input on it.

@Romamo

[Romamo]

Replaced data retrieval to direct read from db and typesense
Added app permission check