Applied mathematics, put to work.

Our first domain is the markets: automated trading driven by AI models, built on mathematics we hold to the standard of a proof. Others will follow.

Approach

Rigorous where it matters, automated where it can be.

Mathematics first

Every model starts as mathematics, worked through under close human scrutiny until it is exactly right. Code and tests follow from it.

Designed by people

The architecture and design of our core systems are human work. AI agents are tools we use.

Finance first

Models trained on market data, held to strict baselines, running on our own data, training and trading stack.

Open source

What we can share, we publish.

Much of our work stays in-house. Tools that are useful beyond it, we release.

render-agent-log

  • GitHub Action
  • Claude Code
  • Apache-2.0

Turns a Claude Code run into a log you can read in the GitHub Actions UI: one line per tool call, highlighting, and fold-away details. It works live while the agent runs, or on a finished run.

A Claude Code run in the GitHub Actions log: a failing test, the files it read, the edit as a diff, the tests passing, and a summary