About Spanlens

The lens you point at every LLM call. Open source, drop-in, self-hostable.

Why we built this

We were shipping LLM apps and burning cash on calls we didn't fully understand. The existing observability tools were either heavy (long onboarding, opinionated SDK you wrap every chain in) or sat on a pivot (acquired into a bigger product, public roadmap quiet). Neither felt right for a solo dev or a small team that just wants to see what their AI is doing.

Spanlens started as a proxy you point at OpenAI to log everything, then grew into tracing, evals, anomaly detection, and a model-savings recommender, all in one MIT-licensed repo with no ee/ folder. The same code we run for hosted customers is the code you self-host. There is no second build.

Our principles

  • One line to integrate. If you cannot turn Spanlens on by changing one line, we have failed.
  • Never on the critical path. p99 ingestion overhead is under 3ms. If Spanlens fails, your request still completes.
  • All features, all plans, all builds. No ee/ folder, no enterprise-only paywall on security features, no telemetry-on- by-default in self-hosted.
  • Numbers, not vibes. Cost savings come with dollar figures. Prompt A/B comes with a Welch t-test. Eval drift comes with a judge-to-human correlation metric.

Who is building this

Spanlens is built by a small team led by Haeseong Kim (founder, engineering). Background in production LLM application development and full-stack engineering. The team has shipped agent systems, RAG pipelines, and proxy infrastructure at scale before starting Spanlens in early 2026.

We are open to contributions. The roadmap, issues, and source live on GitHub. Bug reports, feature requests, and pull requests are all read and answered.

License and stack

Every line of Spanlens ships under the MIT license. The stack is Next.js 14 + Hono + Supabase Postgres + ClickHouse, with TypeScript and Python SDKs. The full stack runs from one Docker compose file on your own infrastructure if you prefer not to use the hosted plan.

Contact

Ready to point the lens at your LLM calls?

Start free →Read the docs