Failproofai
Failproofai is runtime failure resolution for coding agents — catches loops, bad tool calls, and dangerous actions locally with zero latency.
Last verified:
What is Failproofai?
Failproofai (Failproof AI) is an auto-healing plugin for AI coding agents that catches and mitigates runtime failures in production. The harness plugin for agent error handling detects five key failure modes — infinite loops, hallucinated tool/file calls, context drift, intent completion overshoot, and dangerous/destructive actions — using 39 built-in policies that run locally with 0ms added latency and no data leaving your machine. Features include a local dashboard for monitoring sessions and inspecting tool calls, custom policies in JavaScript via an allow/deny/instruct API, three-scope configuration with automatic merging, and convention-based policy auto-discovery. Works with Claude Code, OpenAI Codex, GitHub Copilot, Cursor Agent, Opencode, Gemini CLI, and more.
Failproofai pricing
Pricing model: Freemium
Failproof AI is open source and free to use. The package is installed via npm install -g failproofai or bun add -g failproofai at no cost. All 39 built-in policies are included free. The tool runs locally with no cloud costs. There is no mention of paid tiers or enterprise plans on the website. Teams can book a demo via the 'book a demo' button on the homepage, but the core tool remains free and open source.
Failproofai pros
- 39 built-in policies catch common agent failure modes out of the box
- 0ms added latency - runs as separate process alongside agent
- Local-only execution - no telemetry, no cloud round-trip, data never leaves machine
- First-class support for Claude Code, Codex, Copilot, Cursor, Opencode, PI, Gemini
- Detects and mitigates infinite loops by breaking loop and summarizing progress
- Blocks hallucinations by detecting non-existent tool/file calls
- Prevents context drift by pulling agent back to original goal
- Stops agents from doing destructive actions with human approval option
- Custom policies written in JavaScript with allow/deny/instruct API
- Agent Monitor dashboard to browse sessions and inspect every tool call
- Three-scope configuration (global, project, local) merges automatically
- Convention-based policy auto-discovery - drop files in .failproofai/policies/
- Open source with 32 GitHub stars and active development
- Configurable policy parameters like allowPatterns, protectedBranches, thresholds
- Prevents secret leakage including JWT tokens, API keys, connection strings, private keys
Failproofai cons
- Only supports Node.js >= 20.9.0 (requires recent Node version)
- Currently no support for Goose agents (coming soon only)
- Bun >= 1.3.0 required only for development/building from source
- Telemetry enabled by default via PostHog (must set env var to disable)
- Custom policy errors are fail-open which could allow unintended operations
- Limited to harness-based agents - may not work with non-hook-based agents
- Dashboard only accessible at localhost:8020 (no remote access)
- 32 GitHub stars indicates relatively new/adoption still growing
Frequently asked questions about Failproofai
What is Failproof AI and what problem does it solve?
Failproof AI is an auto-healing plugin for AI coding agents that catches bad LLM tool calls and prevents transient failures - the #1 reason agents fail in production. It detects and mitigates five key failure modes: infinite loops, hallucinations, context drift, intent completion issues, and dangerous actions. The tool runs locally with 0ms latency and provides 39 built-in policies to keep agents reliable, on-task, and running autonomously.
Which AI agents and tools does Failproof AI support?
Failproof AI provides first-class support for Claude Code (Anthropic CLI + SDK), OpenAI Codex (CLI + agent runtime), GitHub Copilot (Copilot CLI hooks), Cursor Agent (Cursor CLI hooks), Opencode (Opencode plugins), PI coding agent (pi-coding-agent), and Gemini CLI (Google Gemini CLI). Goose deep agents are coming soon. The tool plugs into the hooks that the harness already exposes.
How do I install Failproof AI?
Install via npm: npm install -g failproofai, or via Bun: bun add -g failproofai. Then enable policies with failproofai policies --install which writes hook entries into ~/.claude/settings.json. Launch the local dashboard with failproofai which opens http://localhost:8020. Check active policies with failproofai policies.
Does Failproof AI send my agent data to the cloud?
No. Failproof AI is local-only with no telemetry and no cloud round-trip. Your agent's reasoning never leaves your machine. The tool runs as a separate process alongside your agent, reads the trace, and never blocks the loop. Telemetry via PostHog is anonymous usage data only - no session content, file names, tool inputs, or personal information is ever sent.
What are the built-in policies and what do they do?
Failproof AI includes 39 built-in policies such as: block-sudo (prevent privileged commands), block-rm-rf (prevent recursive deletion), block-curl-pipe-sh (prevent piping untrusted scripts), sanitize-jwt/sanitize-api-keys/sanitize-connection-strings (stop credential leakage), block-env-files (prevent reading .env files), block-read-outside-cwd (keep agents in project boundaries), block-push-master/block-force-push (prevent dangerous git operations), warn-destructive-sql/warn-schema-alteration (catch dangerous SQL), warn-large-file-write, warn-package-publish, and warn-background-process. All policies are configurable.
How do I create custom policies?
Write custom policies in JavaScript using the allow/deny/instruct API. Import { customPolicies, allow, deny, instruct } from 'failproofai' and add policies with customPolicies.add({ name, description, match: { events }, fn }). The context object provides eventType, toolName, toolInput, payload, and session info. Install with failproofai policies --install --custom ./my-policies.js. Convention-based policies auto-load from .failproofai/policies/ without the --custom flag.
What is the configuration system and how do scopes work?
Policy configuration lives in ~/.failproofai/policies-config.json (global) or .failproofai/policies-config.json (per-project). Three config scopes merge automatically: project → local → global. Global writes to ~/.claude/settings.json, project writes to .claude/settings.json, and local writes to .claude/settings.local.json. You can tune policies with allowlists, protected branches, and thresholds per-project or globally.
How does Failproof AI add zero latency?
Failproof AI piggybacks on the harness hooks your agent already calls and never sits in the request path. It runs as a separate process alongside your agent, reads the trace, and never blocks the loop. This architecture ensures 0ms added latency while still providing real-time failure detection and mitigation.
How do I disable telemetry?
Failproof AI collects anonymous usage telemetry via PostHog by default. To disable it, set the environment variable FAILPROOFAI_TELEMETRY_DISABLED=1 before running the tool: FAILPROOFAI_TELEMETRY_DISABLED=1 failproofai. No session content, file names, tool inputs, or personal information is ever sent - only anonymous usage data.
What is the 3U Framework for agentic reliability?
The 3U Framework (Uncover, Understand, Utilize) is Failproof AI's model for agentic reliability. In traditional software, retries fix transient errors, but in AI agents, failures are cognitive. Observability after the fact doesn't help when every failure is unique. The framework introduces uncovering failure patterns, understanding them through agent thinking patterns, and utilizing corrective context at runtime to make agents more reliable.