Local-first codebase memory for AI agents

Give your agent searchable memory over your repo.

ragbox builds a semantic snapshot of local code and docs, then lets Codex, Claude, Gemini, or your own scripts ask where things live. Ollama-compatible, terminal-native, and no vector DB server.

Star on GitHub Read the docs

brew tap kroq86/fasm-mac https://github.com/kroq86/fasm-mac
brew install ragbox
brew install ollama
ollama pull nomic-embed-text

arch -x86_64 ragbox build --root . --out memory.lv
arch -x86_64 ragbox search --index memory.lv \
  --query "where is auth handled?" --json

Local files

Indexes are copyable `.lv` snapshots plus JSON manifests. Keep them in your workspace, CI, or release artifacts.

Agent-ready

Use semantic search to feed repo context into coding agents without deploying a web platform.

Fast native core

FASM owns dot/norm/top-k kernels; C++ owns chunking, Ollama glue, mmap search, and the CLI.

Why not the obvious alternatives?

vs ripgrep

ragbox is semantic search, not lexical search.

vs a vector DB

ragbox is a file snapshot, not a running service.

vs a RAG platform

ragbox is a local CLI for repo memory, not a web app.