Spanlens Reviews and Details

Questions & Answers

1. What's the story behind Spanlens?

   I was building LLM apps on the side and kept pasting token counts into a spreadsheet to figure out what each feature cost me. The tools I tried were either acquired mid-migration, closed source, or heavier to self-host than the app I was trying to monitor. So in April 2026 I started building the tool I actually wanted: change one line, see every request. It launched in June 2026, and the whole codebase went up on GitHub under MIT from day one.

2. What makes Spanlens unique?

   The entire product is MIT licensed, including the dashboard, evals, and prompt A/B testing. There is no separate enterprise edition. Everything ships in one repo you can run with a single Docker Compose file. Integration is one line: you change the baseURL on your OpenAI, Anthropic, or Gemini client, and every call gets logged with its full body, token counts, cost, and latency. A few things that are usually paid add-ons come built in, like agent traces with a critical path view, A/B tests that use Welch's t-test to tell you whether a difference is real, and a recommender that flags cheaper models based on the traffic you actually send.

3. Why should a person choose Spanlens over its competitors?

   Mostly because of where the market went. Helicone was acquired, LangSmith is closed source, and self-hosting Langfuse takes real setup work. Spanlens fills the gap those tools left: setup in about five minutes, one Docker Compose file if you want the data on your own servers, and no feature gating between free and paid tiers. To be fair, if you need SOC 2 reports and enterprise support today, the bigger platforms are ahead. If you want request logs, costs, and traces without adopting a heavy platform, that is what Spanlens was built for.

4. How would you describe the primary audience of Spanlens?

   Developers who ship LLM features in production apps. The typical user is a solo developer or a small team that added OpenAI or Anthropic calls to their product and now has no clear picture of what those calls cost or why some are slow. Agent builders are the second group, since multi-step workflows are hard to debug without traces. It is a developer tool through and through: if you don't touch code, you won't get much out of it.

5. Who are some of the biggest customers of Spanlens?

   We don't publish customer names yet. Spanlens launched in June 2026, and most users so far are indie developers and small AI teams.

6. Which are the primary technologies used for building Spanlens?

   TypeScript across the stack. The dashboard is Next.js, the API and proxy run on Hono, and data is split between Supabase Postgres for accounts and relational data and ClickHouse for request logs, which grow fast. The repo is a pnpm monorepo that also holds the JavaScript and Python SDKs, a CLI, and an MCP server. Self-hosting runs on Docker Compose, and OpenTelemetry traces can be ingested over OTLP.