Epic: Intelligent AI Chat System Enhancement

[victorZKov]

Overview

Transform Chivato's basic chat into a production-ready AI-powered assistant with streaming, context-awareness, and actionable suggestions.

Reference Implementation

Based on proven patterns from prl-cc project.

Scope

This epic covers:

1. UI enhancements (display modes: floating, docked, fullscreen)
2. Page context awareness (auto-detect viewed entity)
3. Real-time streaming responses via SignalR
4. Vector embedding pipeline for semantic search (Chivato.Indexer)
5. Proposed actions system (AI-suggested executable commands)
6. Azure OpenAI integration (replace mock service)

Sub-issues

• Chat UI: Add docked and improved fullscreen modes #87 - Chat UI: Add docked and fullscreen modes
• Chat: Page context system with useChatPageContext hook #88 - Page context system with useChatPageContext hook
• Chat: SignalR streaming for AI responses #89 - SignalR streaming for AI responses
• Chivato.Indexer: Vector embedding service for semantic search #90 - Chivato.Indexer service for vector embeddings
• Chat: Proposed actions with execution capability #91 - Proposed actions with execution capability
• Chat: Connect existing Azure OpenAI to chat service #92 - Azure OpenAI provider integration

Architecture Reference

From prl-cc:

• ChatHub.cs for SignalR streaming
• RAGPipeline.cs for context retrieval
• EntityContextService.cs for deep context
• PRL.Indexer for embedding generation
• chatStore.ts and display modes in frontend

Suggested Implementation Order

1. Chat: Connect existing Azure OpenAI to chat service #92 Azure OpenAI provider (foundation)
2. Chat UI: Add docked and improved fullscreen modes #87 UI modes (user-facing improvement)
3. Chat: Page context system with useChatPageContext hook #88 Page context (enhances AI relevance)
4. Chat: SignalR streaming for AI responses #89 SignalR streaming (better UX)
5. Chat: Proposed actions with execution capability #91 Proposed actions (advanced feature)
6. Chivato.Indexer: Vector embedding service for semantic search #90 Indexer service (RAG capability)

Success Criteria

• Chat supports 3 display modes (floating, docked, fullscreen)
• Chat auto-detects when viewing Pipeline, Drift, or Scan
• Responses stream token-by-token
• Semantic search finds relevant historical drifts
• AI can suggest "Resolve Drift", "Ignore", "Rescan"
• Production AI provider (Azure OpenAI) integrated