agentkitai/agentlens

docker run -p 3400:3400 -e AUTH_DISABLED=true -e JWT_SECRET=dev-secret ghcr.io/agentkitai/agentlens
npx @agentkitai/agentlens-server
pip install agentlensai[all-providers]

Usage

• •Point your OTLP exporter at http://localhost:3400/v1/traces to ingest gen_ai.* spans from any OTel-instrumented agent with no AgentLens SDK.
• •Instrument Python in one line: agentlensai.init(url=..., api_key=..., agent_id=...) — captures every LLM call across 9 providers automatically.
• •Add the MCP server to Claude Desktop/Cursor with AGENTLENS_API_URL/AGENTLENS_API_KEY to get 22 MCP tools.
• •Verify audit integrity: GET /api/audit/verify/export gives auditors a signed JSON snapshot; the demo (./demo/aha.sh) shows tamper detection in 30 seconds.

Key features

• ✓Tamper-evident, append-only, cryptographically verifiable hash chain per session — built for EU AI Act Article 12 record-keeping.
• ✓Five integration paths: OpenTelemetry GenAI, OpenClaw plugin, 1-line Python auto-instrumentation, MCP server, and Python/TypeScript SDKs.
• ✓LLM analytics with cost reconstruction from model pricing, 5-dimension health scores, cost optimizer, session replay and A/B benchmarking (Welch's t-test).
• ✓Guardrails with dry-run mode, alerting, multi-tenant isolation with per-tenant data scoping.
• ✓Self-hosted with SQLite by default, no external dependencies; MIT licensed.

Best for

Teams running production agents that need real-time observability plus a compliance-grade, tamper-evident audit trail.

Caveats

• ⚠For anything shared, drop AUTH_DISABLED, set a real JWT_SECRET and create an API key; OTLP can require OTLP_AUTH_TOKEN.
• ⚠Managed AgentLens Cloud (app.agentlens.ai) is a separate paid SaaS alternative to self-hosting.
• ⚠Cost reconstruction from OTel is fuzzy-matched on model id (approximation when no per-call cost attribute exists).
• ⚠MIT licensed.