We build in the open.

Some of the infrastructure behind our AI work is too useful to keep to ourselves. So we release it - MIT-licensed, documented, and maintained. No lock-in, no signup. Take it, fork it, ship it.

Browse the projects → GitHub org

01 - The projects

What we've put out there.

A benchmark you can re-run and a tool that solves one problem well. More on the way.

The twelve all-time winning deliverables of the Startrise LLM Benchmark, one per brief

sr-llm-benchmark

The Startrise LLM Benchmark - harness, briefs and published results

Frontier models build the same twelve real frontends, single-shot: Three.js, WebGL, a WCAG seat map, a brownfield change request, HTML email. Every deliverable is driven in a real browser, judged blind by a three-lab panel and rated blind by a human. The repo is the whole pipeline - generate, gate, judge, score, blind-review UI - plus the briefs and the per-run results: 30 models, 353 builds, 1,059 blind verdicts so far.

Benchmark
Node + Playwright
LLM-as-a-judge
MIT

GitHub →Live leaderboard Browse every build

ping-a-human

Human-in-the-loop for any AI pipeline

An MCP server that lets your AI agent reach a real person on the messaging app they already use - and wait for the reply. Notify them when a job finishes, or ask a question and block until they answer or tap a button.

MCP server
TypeScript
MIT

Website →GitHub npm

02 - Why we open-source

The tools we'd want someone to hand us.

Built from real work

These aren't side projects. They come out of client engagements where we needed the thing to exist - so we made it solid, then made it public.

Yours to keep

MIT-licensed and dependency-light. Run it on your own infra, read every line, fork it if you outgrow us. No platform to get locked into.

A window into how we work

Want to see how Startrise ships software before you hire us? Read the code. Same bar as the work we do under NDA.

Next step

Want this wired into your stack?

hello@startrise.io

ping-a-human is free to run yourself. If you'd rather have us design the human-in-the-loop layer - or any AI automation - into your systems, that's what the lab does. Talk to the AI Labs team →See what the lab builds →