Local-first tools for people who run AI agents on their own terms.
Everything here is MIT, runs on your machine, and talks to no server of ours — no backend, no account, no telemetry, in any of it.
Security scanner for local MCP (Model Context Protocol) configurations — Claude Desktop, Claude Code, Cursor, VS Code, Windsurf. It reports what the configuration file itself gives away: credentials written into it in plain text, servers reached in the clear, launch commands that download and run remote code or resolve a package name anyone could claim, and servers handed a whole filesystem or an unrestricted shell.
It reads the configuration, not the servers. Read-only, no network call while scanning, and it never logs a credential value — only that one exists, and where.
pipx install mcp-config-auditPyPI is the only registry this project is published to. Every release is built from a tagged commit in this organization and pushed over a trusted publisher; there is no release path that does not pass through the repo. A package of this name on any other registry is not ours.
A Grammarly-style writing assistant for Chrome that talks to your LLM, with your API key. Inline underlines, a suggestion card and quick rewrite actions — the interaction model kept, the engine swapped for an endpoint you control. Point it at a local model and nothing leaves the machine at all.
Install from the Chrome Web Store · build from source
Turns an epub into a chaptered audiobook. Gemini TTS does the narration; voice conversion against a single reference clip then anchors the timbre, so the voice stops changing every time the model starts a new generation. Resumable, and it shows the cost before it spends anything.
The code is the small part. FINDINGS.md is the research behind it: engine defects measured in semitones and hertz, the blind listening rounds, and the fixes that improved the numbers while sounding worse.
- Every GitHub Action is pinned to a commit SHA, with Dependabot raising the bumps as reviewable PRs — pinned, but not frozen.
- Releases publish over OIDC trusted publishing. There is no long-lived registry token to leak.
- Secret scanning with push protection, private vulnerability reporting, signed commits, and 2FA required across the organization.