stashbase

StashBase turns local files into searchable context for agents. Open a folder and it prepares text from supported files (PDFs, DOCX, images, audio and video), builds a search index that can find things by meaning rather than only by keyword, and shares that context with Claude, Codex and other MCP clients. The README's flow is local files, prepared text, search index, MCP, agents, and it stresses that your folders stay the source of truth because the index can be rebuilt from them.

Setup is via Homebrew (`brew install --cask liliu-z/stashbase/stashbase`) or installers for macOS 12+ on Apple Silicon and Windows 10+ x64, with a community-supported Linux build for x86_64 Debian 12+ and Ubuntu 22.04+. After you open a folder, a built-in agent chat starts. Semantic search is optional and needs an OpenAI or OpenRouter key under Settings, AI Index; transcription needs a downloaded speech model (Tiny 74 MiB, Base 141 MiB, Small 465 MiB). Without AI Index, exact text search still works, and Settings, MCP connects external AI tools.

The library is opt-in: only folders you open are indexed, and removing a folder clears its index without deleting your files. Embeddings are sent to the provider whose key you configure, so private documents leave your machine if you turn on AI Index. Unsigned builds can trigger macOS or Windows warnings, and the README documents workarounds. The licence is Apache 2.0 and the last push was 2026-10-13. This review is based on the README and repository data; we have not run it ourselves.