Loom
Loom, my open source project: local memory for LLMs with multilingual semantic search, no Docker.
❯ start with part 1Loom: Local LLM Memory with MCP Server and CLI, No Docker
Loom is an open source tool written in Go that turns your files into a memory layer LLMs can query. Inspired by Karpathy's llm-wiki pattern, but with its own architecture: precomputed summaries, BM25 on SQLite, zero Docker. It supports Ollama, Claude, and OpenAI.
Loom Grows Up: Optional Hybrid Semantic Search, Multilingual Retrieval, and an HTTP Server
Loom gets an upgrade: optional hybrid semantic search (BM25 + embeddings) arrives to handle the cases where keyword search alone falls short - paraphrases and multilingual queries. On top of that, there is an HTTP server for integrating Loom into other apps and a multi-corpus mode. Fully backward compatible: if you touch nothing, Loom works exactly as before.
Loom 0.7: When the Answer Is on Page 40 of a PDF
Loom 0.7 fixes its most annoying limit with long documents: instead of the first 8000 characters, the LLM now gets the passage that actually answers the question. Plus a loom_grep MCP tool for exact matches, reading long files in ranges, .loomignore, and a bug that silently turned hybrid search off.