textrawl

I kept losing context. Notes in one app, bookmarks in another, PDFs scattered across folders. When I needed to find something, search always failed me.

Keyword search doesn't understand what you're looking for, it just matches strings. I'd search for "authentication" and miss documents that talked about "login" or "OAuth" or "user sessions."

And when working with AI assistants, I constantly wished they could access my personal knowledge base. Instead, I was copying and pasting context into every conversation.

textrawl started as a personal knowledge base, a way to crawl my own documents and make them searchable with semantic understanding, not just keyword matching.

It crawls directories of Markdown, PDFs, HTML, and text files. Creates embeddings using local or cloud models. Stores everything in PostgreSQL with pgvector for efficient similarity search.

The key insight was making it an MCP server. Now Claude and other AI assistants can search my knowledge base directly. My second brain, available in every conversation.

• Hybrid search. Combines semantic similarity with keyword matching for best results.
• Local-first option. Can run entirely on your machine with Ollama embeddings and SQLite.
• Chunking strategy. Smart document splitting that respects paragraph boundaries and maintains context.

• Embeddings are powerful. The ability to find conceptually related content, not just keyword matches, is transformative.
• MCP is the right abstraction. Building as an MCP server means any compatible AI assistant can use it, not just one provider.
• Privacy matters. Local-first architecture lets you keep your knowledge private while still benefiting from AI.