Changelog

All notable changes to the MLCopilot platform will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[1.1.0] - 2026-08-11

GitHub Repository Integration: Support for code-aware intelligence, repository context ingestion, and code file parsing.
Production Google Gemini Integration: Configured gemini-3.6-flash REST streaming provider for cloud production deployments.
Custom Brand Identity: Updated brand monogram, header logo, and transparent browser favicon across dark/light theme engines.
Vercel Production Deployment: Deployed production web application to Vercel connected to Neon PostgreSQL + pgvector cloud database.

Fixed Docker container build verification in CI pipeline (apps/api/Dockerfile).
Resolved API deprecation 404 handling in LLM provider REST calls.

Confidence Routing Engine: Implemented rag_similarity_threshold (0.35) check in RAGService to dynamically route queries between Grounded Workspace RAG and General Conversational Mode.
Ollama Keep-Alive & Connection Pooling: Added persistent httpx.AsyncClient connection pooling and "keep_alive": "15m" payload options in OllamaProvider.
Dynamic Model Candidate Resolution: Auto-selects preference candidate (qwen2.5:3b → llama3.2:3b → llama3.1:8b) via /api/tags.
ChatGPT / Claude Composer Redesign: Floating glassmorphism composer container with smooth focus glow, shortcut pills (@, /), and circular send button.
Optimistic Conversation Deletion: Instant sidebar purge in 0ms using React Query onMutate cache manipulation in useChat.ts.
Multi-Phase Progress Indicators: UI streaming badges displaying Searching workspace... → Thinking... → Generating response....
CI/CD Pipeline: GitHub Actions workflow .github/workflows/ci.yml for automated linting, backend tests, frontend builds, and Docker image validation.

Refined system prompt in PromptBuilder to adhere to 6 explicit operating rules for partial questions, workspace missing data, and citations.
Reduced default context window from 4 to 3 top-scoring deduplicated chunks to reduce token footprint.
Scoped citations block to render strictly when RAG citations exist.

Fixed UI blink on SSE stream completion by awaiting React Query invalidation before resetting stream state.
Fixed stale sidebar state during conversation deletion.
Fixed hardcoded model labels by connecting frontend dynamically to /api/v1/health/llm.

Initial release candidate of MLCopilot platform featuring PostgreSQL pgvector, Next.js App Router, FastAPI backend, and document parsing engine (PDF, DOCX, TXT).