🔍 AgentForensics: The Flight Recorder for AI Agents. When They Crash, You Know Exactly Why.

"95% of AI pilots fail to reach production. 34% of firms trust their agents. When your agent fails in production, you have no idea why. Until now."

AgentForensics is the missing runtime forensics and production debugging layer for your AI agents. It provides a full root cause analysis, agent replay, and deep agent observability so you can stop guessing and start fixing.

This is the ultimate AI agent debugging tool.

🚀 The 5-Second Quickstart

pip install agentforensics
curl -sSL https://raw.githubusercontent.com/ASHOKMUKHERJEE6/agent-forensics/main/scripts/install.sh | bash

Then in your agent code:

from agentforensics import trace

@trace
def my_agent(task):
    # Your agent code here
    pass

🏗️ Architecture

graph LR
    A[Agent] -->|Trace Recorder| B[Ingestion]
    B -->|Batch Write| C[(ClickHouse)]
    C -->|Real-time Poll| D[Anomaly Detector]
    D -->|Alert| E[Alert System]
    C -->|Query| F[Replay Engine]
    F -->|Render| G[Debugger UI]

Loading

🕹️ The Agent Trilogy

AgentAutopilot	AgentQA	AgentForensics
Stops agents from burning money	Stops agents from shipping garbage	Helps you debug when they still fail
Runtime circuit breaker	Pre-merge quality gate	Post-mortem flight recorder
Reactive (kills)	Proactive (prevents)	Diagnostic (explains)
Focus: Cost	Focus: Quality	Focus: Reliability

Autopilot protects your wallet. QA protects your reputation. Forensics protects your sanity.

🤝 Contributing

Please see CONTRIBUTING.md for details on our PR templates and guidelines.

📝 License

Apache 2.0 License. See LICENSE for more information.