Hey hunters 👋 I'm the maker of Fida. I built this because I use AI coding agents every day, and I kept noticing how casually they pull .env files, config, and command output straight into model context. Your secrets end up somewhere you never see — and existing tools either scan git history after the leak, or wrap everything in a heavy policy firewall so annoying that people just turn it off. Fida takes a narrower stance: a detected secret must never reach the model, full stop. When an agent needs sensitive content, Fida hands it a redacted view (API_KEY=[REDACTED]) — but the actual command still runs with the real credential, so nothing breaks. The agent uses the secret without seeing it. A few choices I care about: • Local-first — nothing leaves your machine. • Fail-closed — if Fida can't prove a response is clean, it suppresses it rather than leak a fragment. • Honest scope — it names its coverage (enforced / best_effort) instead of pretending to protect everything. It's open-source (Rust, MIT) and install-and-forget — one command wires every coding agent it detects (Codex, Claude Code, Cursor, Windsurf, and more). Would love your feedback, especially on detector coverage and which agents you'd want supported next. 🙏

Fida

Let agents read your code, not your secrets.

Let agents read your code, not your secrets.