Doc Agent

# Roadmap ## Phase 1: Foundation 🏗️ ✅ - [x] Project setup with `pnpm` workspaces. - [x] TypeScript configuration with strict ESM support (`NodeNext`). - [x] Core package architecture (`core`, `extract`, `vector-store`, `cli`). - [x] Basic Gemini extraction logic. - [x] MCP Server shell. ## Phase 2: Core Capabilities 🚀 (In Progress) ### Ingestion Pipeline ✅ (Epic #1) - [x] SQLite persistence layer with Drizzle ORM. - [x] Ollama extraction strategy (privacy-first default). - [x] Auto-install Ollama via Homebrew. - [x] Auto-pull models with progress bar. - [x] OCR preprocessing with Tesseract.js for accuracy. - [x] Zod validation for AI responses. ### Enhanced Extraction ✅ - [x] **Multi-modal Support**: PDFs and images (PNG, JPEG, WebP). - [x] **Provider Expansion**: - [x] Ollama (Local LLM) - default - [x] Gemini (Cloud) - [ ] OpenAI ### Vector Search Integration (Next) - [ ] Implement `VectorStore` using **LanceDB**. - [ ] Create `index` command to ingest documents. - [ ] Create `search` command for semantic queries. - [ ] Expose search via MCP tool `search_documents`. ## Phase 3: Developer Experience & Polish 💅 (Partial) - [x] Ink-based CLI with rich interactive output. - [x] Progress bars and streaming output. - [x] Comprehensive test coverage (~75%). - [ ] End-to-End tests with real PDFs. - [ ] Documentation for setting up local development. ## Phase 4: Future 🔮 - [ ] Recursive document chunking for large files. - [ ] Watch mode for auto-indexing folder changes. - [ ] Additional local models (Deepseek, etc.).