DEV Community

Jonathan Martin Paez
Jonathan Martin Paez

Posted on

Local-first observability for AI agents, in one command

#ai #opensource #typescript #observability

When an AI agent misbehaves — fails, stalls, or quietly burns tokens — you need
to see the steps. But most observability tools are cloud-first: accounts, API
keys, and shipping your prompts to someone else's servers.

I built Lookspan to be the opposite. It runs on your machine, stores everything
in local SQLite, and starts with one command:

npx lookspan
Enter fullscreen mode Exit fullscreen mode

Open http://127.0.0.1:3100 and you get a real-time dashboard: traces, a span
graph, cost per model, latency percentiles, and alerts.

Send it data

Any language, raw HTTP:

curl -X POST http://127.0.0.1:3100/api/ingest -H "Content-Type: application/json" -d '{...}'
Enter fullscreen mode Exit fullscreen mode

MCP (TypeScript):

npm i @lookspan/mcp
Enter fullscreen mode Exit fullscreen mode

Python (LangGraph, CrewAI, or generic):

pip install lookspan
Enter fullscreen mode Exit fullscreen mode

Already using OpenTelemetry? Point your OTLP exporter at it — no Lookspan SDK:

OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=http://127.0.0.1:3100/v1/traces
Enter fullscreen mode Exit fullscreen mode

[aquí: GIF de demo + captura del dashboard]

It's MIT and early (v0.1). Repo + roadmap: https://github.com/JoniMartin27/lookspan

Top comments (0)