Repolode
Submit
FailproofAI

FailproofAI/failproofai

Observability and policy enforcement for AI coding agents and assistants, blocking dangerous tool calls before they run across 12 harnesses.

View on GitHubWebsite
Stars
6,000
Forks
506
Open issues
61
Last commit
yesterday

What is it?

Failproof AI is a local tool that records what AI agents do and can block risky actions before they happen. It hooks into 12 agent harnesses, including coding CLIs such as Claude Code, Codex, OpenCode and Goose, and chat or assistant gateways such as Hermes and OpenClaw. Each run is captured into a session history that you can browse in a local dashboard served on localhost:8020, with no account required.

Enforcement works through policies. The project ships 40 built-in policies, such as blocking reads of .env files, recursive deletes, sudo, force pushes to git, and destructive SQL, and teams can add their own policies as code. Each policy can allow an operation, deny it with a message back to the agent, or let it through with added context. Agents that are not covered by a harness can report runs through a Python SDK for tracing and sessions.

The project also includes an offline audit command that scans session history for risky patterns and suggests policies to address them. A hosted Failproof AI Observability product builds on the same data model for fleet-wide monitoring, but the local tooling runs entirely on your machine.

Key features

  • 01Hooks into 12 agent harnesses, including Claude Code, Codex and Hermes.
  • 02Blocks risky tool calls such as .env reads, sudo, and force pushes before they execute.
  • 03Ships 40 built-in policies and supports custom policies written in JavaScript or TypeScript.
  • 04Runs a local dashboard that shows sessions, model calls, tool calls and hook decisions.
  • 05Provides an offline audit that scans history for risky patterns and suggests policies.

Use cases

  • Preventing a coding agent from deleting files, force-pushing to main, or running destructive SQL.
  • Keeping secret files like .env out of an agent's context.
  • Reviewing what an agent did in past sessions and finding recurring failure patterns.
  • Sharing team-wide guardrails by committing policy files to a repository.

Topics

#ai-agent#observability#guardrails#claude-code#llm-observability#agent-reliability#policy-enforcement#agent-failure#agent-harness#agent-observability

FAQ

What is failproofai?
Observability and policy enforcement for AI coding agents and assistants, blocking dangerous tool calls before they run across 12 harnesses. Failproof AI is a local tool that records what AI agents do and can block risky actions before they happen. It hooks into 12 agent harnesses, including coding CLIs such as Claude Code, Codex, OpenCode and Goose, and chat or assistant gateways such as Hermes and OpenClaw. Each run is captured into a session history that you can browse in a local dashboard served on localhost:8020, with no account required.
Is failproofai free to use? What license does it have?
failproofai is open source under the Other license. Check the license terms before using it in commercial products.
Is failproofai still maintained?
Actively maintained. The last commit to failproofai was on Oct 21, 2026.
What can I use failproofai for?
Preventing a coding agent from deleting files, force-pushing to main, or running destructive SQL. Keeping secret files like .env out of an agent's context. Reviewing what an agent did in past sessions and finding recurring failure patterns. Sharing team-wide guardrails by committing policy files to a repository.
Where can I get failproofai?
The source code and installation instructions are on GitHub: https://github.com/FailproofAI/failproofai

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects