The Harden AI Firewall is a local-first AI firewall for coding and productivity agents. If AIF is useful to you please star this repo.
curl -fsSL https://aif.harden.run/install.sh | shThen:
aif configure
aif demo
aif show
aif show --tui
| Agent | Coverage |
|---|---|
| Claude Code | native hooks, pre-execution blocking |
| Codex | native hooks, pre-execution blocking |
| Cursor (IDE + CLI) | hooks, pre-execution blocking |
| Gemini CLI | hooks, pre-execution blocking |
| Kiro | hooks for the main agent; subagent MCP calls via proxy |
| OpenClaw, Hermes | block-and-steer via a bridge process |
See more at our trust page.
We never collect code, commands, file paths, prompts, or anything a secret could ride in. We collect anonymous operational counters only: version, OS/arch, adapters in use, decision/block counts.
aif telemetry showprints the exact payload before it is sent — a secret-safety tool inspecting its own egress.- Opt out with one environment variable:
AIF_NO_TELEMETRY=1.
For full schema see trust.
Raise an issue with this attachment
aif doctor --bundle > aif-diagnostics.txtYou can use these channels
- Email security@harden.run, or
- Open a private report via this repository's Security → Report a vulnerability (GitHub Security Advisories).
Include the aif doctor --bundle output.