Docs

You need an AgentLens workspace and Node 20.9 or newer. If you don't have a workspace yet, create one — the Solo plan is free and needs no card.

Create a collector token in AgentLens under Settings → Collector tokens, then run the collector on the machine your agents run on:

npx @sinaispace/agentlens connect --token al_...

This detects the agents installed locally, verifies the token against the server, writes ~/.agentlens/config.json with mode 600, and backfills the history it finds. Useful flags:

• --url — point at a self-hosted install instead of the cloud.
• --redact — upload metrics only and leave all prompt and response text on the machine.
• --no-sync — connect without the initial backfill.

npx @sinaispace/agentlens sync      # upload anything new since last sync
npx @sinaispace/agentlens status    # what is connected, how far it has synced

Re-syncing is always safe. The server deduplicates on session and event identity, so --full re-reads everything without creating duplicates. Use --dry-run to preview a batch and --max-sessions <n> to cap a large first backfill.

Agents that write session logs locally are detected automatically by the collector:

• Claude Code (Anthropic) — Local session logs + OTel hooks
• Codex CLI (OpenAI) — Session rollouts + API usage
• Gemini CLI (Google) — Telemetry export + local history
• OpenCode (SST) — Storage dir + server events
• Grok CLI (xAI) — API usage + local transcripts · beta
• Qwen Code (Alibaba) — Local history + DashScope usage · beta
• Aider (Open source) — Chat history + analytics file
• Cline (Open source) — Task store + provider usage · beta
• Continue (Open source) — Dev data events · soon

Agents that only report at organisation level can't be seen from a developer machine. An admin connects those once in the web app with org credentials:

• GitHub Copilot (GitHub) — Copilot metrics API + CLI logs
• Cursor (Anysphere) — Admin usage API · beta
• Windsurf (Codeium) — Teams analytics API · soon
• Amp (Sourcegraph) — Thread export API · soon

Anything else can push straight to the ingest API.

The collector reads these, in preference to its config file:

• AGENTLENS_TOKEN — collector token, overrides the config file.
• AGENTLENS_URL — server URL, for self-hosted installs.
• AGENTLENS_CONFIG — config path, default ~/.agentlens/config.json.
• CLAUDE_CONFIG_DIR — Claude Code data directory, default ~/.claude.

The collector is MIT licensed and has no runtime dependencies — read it on GitHub or on npm. It is read-only by design: no proxy in front of your model calls, no provider API keys read or transmitted, and if AgentLens is unreachable your agents keep working and the collector backfills next run.