KaGty1 / AITTAK

What is AITTAK?

AITTAK is a Python-based AI proxy platform designed for offensive security testing ("red team" operations). It sits between your application and upstream AI providers like Claude and OpenAI, intercepting every request and response. The platform logs all prompts, tool calls, and responses to a local SQLite database, giving security teams visibility into how AI tools are being used. It also includes built-in pattern matching to detect sensitive data leakage -- think API keys, JWTs, AWS credentials, national IDs, and bank card numbers -- flagged automatically in audit logs. The most distinctive feature is SSE-based tool injection: it can inject arbitrary tool calls into streaming responses, capture the results, and log them for analysis.

Why is it gaining traction?

Security teams testing AI-integrated applications need to understand exactly what data leaves their environment. AITTAK provides a self-hosted proxy that gives full audit trails without relying on third-party logging services. The built-in sensitive data rules cover a surprisingly wide range -- from Chinese national IDs to AWS access keys to PEM private keys. The tool injection capability is particularly valuable for red teamers evaluating AI tool integrations, as it simulates how a compromised AI assistant might be manipulated to exfiltrate data or execute unauthorized actions.

Who should use this?

Red teamers and security researchers evaluating AI integrations in their organization. SOC analysts hunting for credential leakage through AI tools. DevSecOps teams building compliance controls around AI API usage. If you're not in offensive security or security auditing, this is not the tool you're looking for.

Verdict

With an extremely low credibility score and only 18 stars, this project is extremely early-stage. The default admin password is literally "changeme," there's no HTTPS enforcement, and the documentation appears minimal. That said, the core functionality -- proxying, auditing, and tool injection -- is all there and working. If you're a security professional, the source code is readable enough to audit and deploy in a controlled environment. For everyone else, wait for a more mature release with proper hardening and documentation. This is a proof-of-concept that deserves watching, not betting on yet.