brexhq

brexhq / CrabTrap

Public

An LLM-as-a-judge HTTP proxy to secure agents in production

www.brex.comcrabtrap
46
5
100% credibility
Found Apr 21, 2026 at 46 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Go
AI Summary

CrabTrap is a proxy that intercepts outbound HTTP requests from AI agents, evaluates them against static rules or an LLM policy judge, and logs all decisions for auditing via a web UI.

How It Works

1
🔍 Discover CrabTrap

You hear about CrabTrap, a helpful safety net that watches what your AI helper tries to do online and only lets safe actions through.

2
🚀 Get it running

Download and start it easily with a simple command that sets up everything you need, including a log to track activity.

3
🔗 Point your AI to it

Tell your AI assistant to route all its internet requests through CrabTrap by setting a simple address it uses.

4
🛡️ Set your safety rules

Open the friendly web dashboard to create easy rules like 'allow reading info but block changes' or let an AI helper suggest them based on what your assistant does.

5
📊 Watch the action

Check the dashboard anytime to see a full record of requests your AI made, what was allowed or blocked, and why.

✅ Your AI is safely guarded

Now your AI helper works freely but stays protected, with everything logged so you always know what's happening.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 46 to 46 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is CrabTrap?

CrabTrap is a Go-built HTTP/HTTPS proxy that acts as a guardrail for AI agents making outbound calls to external APIs like Slack, GitHub, or Gmail. It evaluates every request first against fast static rules (prefix/glob matching), then an LLM-as-a-judge for nuanced policies, forwarding safe ones while blocking others with explanations—all logged to PostgreSQL. Spin it up with Docker Compose for a proxy on localhost:8080 and a web UI on :8081 to manage users, policies, and audits.

Why is it gaining traction?

Unlike basic API gateways, CrabTrap combines deterministic rules with LLM-as-a-judge for agent-specific security, plus SSRF blocking, per-IP rate limits, and TLS interception via dynamic certs. Its eval system replays audit logs to test policy accuracy (check awesome llm as a judge github for evals), and an agentic policy builder auto-drafts rules from traffic patterns. Brex open-sourced it for production agents, drawing eyes in llm as a judge evaluation github circles.

Who should use this?

Security engineers securing prod AI agents that hit third-party APIs, without human approvals or response filtering. DevOps teams at fintechs or startups needing outbound guardrails for tools like CrabTrap in Go HTTP setups. Anyone evaluating llm-as-a-judge frameworks github for agent proxies.

Verdict

Grab it if you're shipping agents today—solid quickstart, web UI, and evals make it practical despite 46 stars and 1.0% credibility score signaling early days. Docs shine, tests are thorough; watch for community growth before heavy prod reliance.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.