Launching today
Spanlens
Open-source LLM observability in one line of code
7 followers
Open-source LLM observability in one line of code
7 followers
Spanlens records every OpenAI, Anthropic, Gemini, Azure, and Ollama call with one line of code. Get cost, latency, agent traces with Critical Path, anomaly alerts, PII scan, and model-swap suggestions out of the box. MIT, self-hostable, free tier that doesn't punish growth.
Hey Product Hunt,
I'm Haeseong, the maker of Spanlens.
Most LLM observability tools ask you to restructure your code before you see anything useful. Spanlens doesn't. You swap your OpenAI client for createOpenAI from @spanlens/sdk and you're recording traces. Python works the same way with pip install "spanlens[openai]".
That's the whole setup.
What you get out of the box:
Cost, latency, and full agent span trees
Prompt version tracking and PII scanning
Critical Path analysis (which span actually dominated wall-clock time)
LangGraph topology view on trace detail pages
Welch t-test on the compare page for statistical significance
Works across OpenAI, Anthropic, Gemini, Azure OpenAI, Ollama, Vercel AI SDK, LangChain, and LlamaIndex
How we run it:
MIT licensed, one Docker image to self-host
Free cloud tier with a hard 429 at 50K requests per month, so a runaway dev loop can't bill you
Public changelog with Atom feed, plus a public status page
To be honest, we're early. If you need a mature eval framework with 30+ integrations today, other tools fit better. But if you want the core 80% to just work and you're tired of instrumenting before you can observe, give us 30 seconds.
Repo: github.com/spanlens/Spanlens
Docs: spanlens.io/docs
Changelog: spanlens.io/changelog
I'll reply to every comment today.
Haeseong