Feat/chat streaming

[Haile-12]

Pull Request

Description

i applied those changes

Backend

LLM Client-->llm_clients.py

• Added streaming with retry and fallback using LangChain for Gemini and OpenAI.

RAG Generator-->rag_generator.py

• Added streaming RAG generation that yields partial token deltas and a final formatted response.

Chat Endpoint--> chat.py

• Replaced single-response POST chat handler with SSE streaming endpoint using an event generator.

EventGenerator--> event_generator.py

• Added new class-based EventGenerator

Frontend

MessageList.ts

• Scroll once on each new message.
• Stop auto-scroll if the user scrolls up during streaming.
• Continue following assistant tokens only while streaming and if the user is at the bottom.

Chat.tsx

• Pass isSendingMessage down to MessageList as isStreaming to manage scroll during streaming.

MessageBubble.tsx

• Show the “Thinking…” animation only for the initial placeholder, not during streamed text arrival.

chatService.ts

• Added a streaming client for /api/chat that reads SSE chunks over fetch.

useChatStore.ts

• Made sendMessage streaming-first, updating the thinking bubble in real-time as chunks arrive.
• Implement fallback to the previous non-streaming request.

Checklist

• I have read the contributing guidelines
• I have added or updated tests where applicable
• I have added or updated documentation where applicable
• I have reviewed and tested my changes, and I have run the code to confirm it works properly

How Has This Been Tested?

The changes were tested end-to-end by sending messages through the chat to verify that streaming responses display correctly with partial token updates, updated UI elements such as auto-scroll and the “Thinking…” animation behave as intended, and the streaming chat functionality works reliably alongside existing features.

Additional Notes

Add any additional context or information about the pull request here.

[birukabza]

• u can stream without importing specific model library, utilize the already exisiting init_chat_model. refer here

[birukabza]

• the file location for this file is not logical.
• make handle post streaming a background task

[birukabza]

-ensure the generator is wrapped in a try finally block that guarantees the message is saved even if the connection drops.