When an AI agent starts burning tokens, stalling mid-task, or silently failing in production, you need visibility into every step of its execution. That's exactly the problem JoniMartin27 set out to solve with Lookspan — a local-first observability platform purpose-built for AI agents that requires zero accounts, no API keys, and gets running with a single terminal command.

One Command to Full Visibility

Running npx lookspan spins up the entire stack on your local machine. Open http://127.0.0.1:3100 in your browser and you're greeted with a real-time dashboard featuring traces, a span graph visualization, per-model cost breakdowns, latency percentiles, and configurable alerts. Everything stays on your hardware — no data leaves your environment unless you explicitly push it somewhere.

Language-Agnostic by Design

Lookspan doesn't lock you into any particular stack. The platform exposes a raw HTTP ingestion endpoint at /api/ingest for maximum flexibility: curl -X POST http://127.0.0.1:3100/api/ingest -H "Content-Type: application/json" -d '{...}'. For TypeScript developers, an official MCP SDK (@lookspan/mcp) is available via npm. Python users working with LangGraph, CrewAI, or custom agent frameworks can install the package directly with pip install lookspan — no vendor lock-in, just straightforward instrumentation.

OpenTelemetry Natively

Already using OTEL collectors in your pipeline? You don't need to adopt a Lookspan-specific SDK at all. Point your existing OTLP exporter at http://127.0.0.1:3100/v1/traces and keep your current instrumentation intact. This makes incremental adoption trivial for teams already invested in the OpenTelemetry ecosystem — swap out the endpoint, get instant local observability without rewriting spans or trace contexts.

Early Days, Big Potential

The project is MIT-licensed and currently at v0.1, meaning it's fresh but accessible to contributors and early adopters who want to shape its direction. The GitHub repository (github.com/JoniMartin27/lookspan) hosts the codebase alongside a public roadmap for upcoming features. For developers tired of cloud-first observability tools that demand credit cards just to peek at your own agent traces, Lookspan offers a compelling alternative built on developer-friendly principles: run locally, store in SQLite, and pay nothing.

Key Takeaways

  • Single npx lookspan command launches complete local observability stack with dashboard at port 3100
  • Supports HTTP ingestion, TypeScript MCP SDK, Python packages for LangGraph/CrewAI, and native OpenTelemetry OTLP export
  • All telemetry data stored locally in SQLite — no third-party servers involved
  • MIT licensed, v0.1 stage, with public roadmap on GitHub for community input

The Bottom Line

Lookspan fills a real gap for developers debugging AI agents without surrendering their prompts or trace data to SaaS platforms. It's refreshing to see tooling that respects developer privacy by default rather than treating your debugging sessions as product inventory. Worth watching as it matures — and even more worth trying now if you've ever wanted observability that doesn't require a billing department sign-off.