Show HN: AgentLens – Open-source observability and audit trail for AI agents

1 point

4 months ago

Built AgentLens to solve a problem we kept hitting: once you deploy AI agents, you have no idea what they're actually doing. No traces, no costs, no audit trail.

AgentLens is an open-source observability platform for AI agents. It captures every tool call, LLM interaction, and decision as tamper-evident events with hash-chain integrity.

Key features: - MCP-native integration (Model Context Protocol) - Tamper-evident audit trail with SHA-256 hash chains - Cost tracking across models and sessions - Real-time SSE streaming dashboard - Alerting engine (error rate, latency, cost thresholds) - TypeScript SDK for custom integrations

Stack: TypeScript, Hono, React, Drizzle ORM, SQLite, MCP SDK.

5 npm packages under @agentlensai (core, server, mcp, sdk, cli). 413 tests. MIT licensed.

Part of a broader ecosystem with AgentGate (human-in-the-loop approvals) and FormBridge (structured data collection for agents).

Would love feedback on the architecture and what observability features matter most to you.