Chrome DevTools for AI Agents — Self-hosted, open-source observability with AI-powered failure analysis

When your AI agent fails in production, you're blind. Logs show API calls, but they don't show you WHY the agent chose tool A over tool B, or WHERE its reasoning went wrong.

AgentLens records every step of your agent's execution and lets you replay, debug, and analyze failures — with AI.

pip install agentlens-observe
docker compose up  # Dashboard at http://localhost:3000

import agentlens

agentlens.configure(server_url="http://localhost:8000", api_key="your-key")

@agentlens.trace
def my_agent(query):
    with agentlens.span("search") as s:
        results = search(query)
        s.set_output(str(results))
    return results

That's it. Your traces appear in the dashboard instantly.

Click "Autopsy" on any failed trace — AI identifies the root cause and suggests a fix. BYO API key (OpenAI, Anthropic, or Gemini).

First-class support for Model Context Protocol. Auto-trace tools/call, resources/read, and prompts/get with zero config.

from agentlens.integrations.mcp import patch_mcp
patch_mcp()  # That's it — all MCP calls are traced

Time-travel through your agent's execution step-by-step. Edit inputs at any span, save sessions, compare original vs modified.

Define scoring criteria, run automated evaluations against traces. Numeric (1-5) or pass/fail scoring with custom rubrics.

Version-control your prompt templates. Compare versions with unified diff. Track which prompt version produced which results.

Extend with SDK SpanProcessors (on_start/on_end hooks) and server-side auto-discovered plugins.

Rule-based alerts on cost, latency, error rate. Webhook notifications when thresholds are exceeded.

LangChain, CrewAI, AutoGen, LlamaIndex, Google ADK, MCP, Semantic Kernel (stub)

sdk/                # Python SDK (PyPI: agentlens-observe)
sdk-ts/             # TypeScript SDK (npm: agentlens-observe)
sdk-dotnet/         # .NET SDK (NuGet: AgentLens.Observe)
server/             # FastAPI backend (Python)
dashboard/          # React web UI (TypeScript)
cli/                # Go CLI tool
vscode-extension/   # VS Code extension

# Server (279+ tests)
cd server && pip install -r requirements.txt
uvicorn main:app --reload --port 8000
pytest

# Dashboard
cd dashboard && npm install && npm run dev

# Python SDK
cd sdk && pip install -e ".[dev]" && pytest

# .NET SDK (29 tests)
cd sdk-dotnet && dotnet build && dotnet test

# Go CLI
cd cli && go build -o agentlens .

Contributions welcome! See issues for good first issues.

See LICENSE for details.