lukehebe

lukehebe / Agent-Drift

Public

IDS for AI Agents.

37
3
100% credibility
Found Feb 03, 2026 at 22 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

Agent Drift provides real-time monitoring for AI agents to detect behavioral changes, prompt injections, and anomalies through a dashboard.

How It Works

1
📖 Discover Agent Drift

You hear about a smart safety watch that keeps an eye on your AI helpers to catch sneaky changes in how they act.

2
💻 Set it up simply

You add this safety watch to your computer in just a couple of minutes, like installing a helpful app.

3
🔗 Connect your AI helper

You link the watch to your AI agent with one easy step, so it can quietly follow along.

4
🖥️ Launch the dashboard

You open a colorful screen that shows everything happening in real time, like a security control room.

5
🤖 Run your AI tasks

Your AI helper does its usual jobs while the watch tracks tools it uses and spots anything odd.

6
🚨 Get smart alerts

If the AI starts acting strangely—like using new tools or slowing down—you see instant warnings with details.

🛡️ AI stays safe

You now have full confidence your AI helper is protected from hidden tricks, running smoothly every time.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 22 to 37 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 Agent-Drift?

Agent-Drift delivers IDS for AI agents, spotting agent drift from prompt injection, memory poisoning, or tool abuse before outputs go haywire. In Python, it baselines normal tool sequences, timings, and decisions via HTTP POSTs to /tool or CLI integration, then flags anomalies on a real-time WebSocket dashboard at localhost:5001. Think SIEM for autonomous systems, with honeypot tools and OWASP Top 10 LLM scans baked in.

Why is it gaining traction?

Zero-config baseline learning and one-command OpenClaw patching hook devs instantly, while honeypots trigger CRITICAL alerts on decoy tool calls. The dashboard's live drift timelines, alert feeds, and API for /check scans stand out over basic content filters—users get forensic traces without wiring deep into agent loops. Drift control agent features like N-gram forbidden sequences catch subtle behavioral shifts alternatives miss.

Who should use this?

AI builders deploying drift ai chat agents or drift chat agents in LangChain, AutoGPT, or CrewAI. Security ops monitoring ice agents ids or ids ips agents for OWASP threats in production. Teams needing anti drift agent for tool-heavy workflows like code exec or file access.

Verdict

Worth a spin for agent security today—CLI like `agent-drift honeypot add` and docs shine, but 1.0% credibility and 34 stars mean alpha risks; test thoroughly before prod.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.