Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

862 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

🛡️ Agentic-Eval: Enterprise AI Agent Security & Observability

Datadog + OWASP Security Guard for Autonomous AI Agents


⚡ Executive Summary

Agentic-Eval is a framework-agnostic, developer-native security and observability platform designed to protect engineering teams from credential leaks, infinite token loops, prompt injections, and unhandled tool crashes in production AI agents.

By combining an active CI/CD security gate (GitHub Action + PyPI CLI) with a sub-millisecond Golang security daemon, Agentic-Eval prevents security scandals and runaway API bills before code reaches production.


🛑 The Core Problem: The 3 Production AI Agent Nightmares

  1. OWASP LLM02 Secret Leaks: Agents accidentally logging OpenAI keys (sk-proj-...), GitHub tokens (ghp_...), or AWS credentials (AKIA...) into public chat windows or database logs.
  2. OWASP LLM08 Infinite Loop Token Burn: Agents getting stuck repeating identical tool calls in a loop, burning $500 - $2,000 in API credits overnight.
  3. OWASP LLM01 Prompt Injection & Excessive Agency: Untrusted web/document data overriding agent instructions and executing unauthorized tool actions.

🎯 Our Technical Architecture & Solution

┌─────────────────────────────────────────────────────────────┐
│  GITHUB ACTION / PYPI CLI (Pre-Commit & CI/CD Pipeline)      │
│  - Blocks vulnerable Pull Requests before production        │
└──────────────────────────────┬──────────────────────────────┘
                               │ High-Speed JSON Inspection
                               ▼
┌─────────────────────────────────────────────────────────────┐
│  GOLANG SECURITY DAEMON (go_services/main.go)                │
│  - Sub-millisecond (<1ms) OWASP Top 10 Pattern Inspection    │
└──────────────────────────────┬──────────────────────────────┘
                               │ REST API Verification
                               ▼
┌─────────────────────────────────────────────────────────────┐
│  FASTAPI ENTERPRISE ENGINE (python_backend/app/main.py)      │
│  - Trajectory Audit Engine + Real-time Token Budget Guard    │
│  - Automated Code Remediation Patch Generator                │
└─────────────────────────────────────────────────────────────┘

🛠️ Key Product Features

Feature Description Market Advantage
Sub-Millisecond Go Proxy Native Go engine scanning trajectories in <1ms. Zero latency penalty vs slow Python tools.
Pre-Commit CI/CD Gate Embeds into .github/workflows/ci.yml in 3 lines. Catches bugs pre-deployment vs post-mortem logs.
Automated Remediation Patches Generates Python/Go code patches for loop breakers & regex scrubbers. Provides direct code fixes, not just passive graphs.
Token Budget Kill-Switch Halts execution if a task exceeds custom USD budget caps. Prevents runaway $1,000+ API credit bills.
SOC2 / OWASP Compliance Exports cryptographic audit reports for enterprise procurement. Enables dev agencies to prove quality to clients.

💼 Business Model & Pricing Tiers

  • Developer API Subscription ($19/mo): Secret scrubbing proxy + trajectory linting API.
  • B2B Trajectory Audit Report ($250/report): Complete OWASP Top 10 audit certificate for client handoffs.
  • Custom Code Remediation Patch ($750/patch): Custom FastAPI / LangChain loop-breaker & secrets patch.
  • Enterprise On-Premises Retainer ($2,500/mo): Air-gapped VPC deployment & dedicated SLA support.

🧪 Verified Quality Assurance

  • 100% Passing Automated Tests: Native Go test suite (go test -v) + Python OWASP security test suite (python test_agentic_eval_security.py).
  • Hosting Footprint: $0.00/mo (Runs on Vercel & Render free tier for maximum profit margins).

About

Enterprise AI Agent Security, Observability & OWASP Top 10 Trajectory Audit Engine (FastAPI + Golang + React)

Resources

Stars

0 stars

Watchers

1 watching

Forks

Releases

Packages

Contributors

Languages