Open source

We turn the maths into tools anyone can run and check. Everything here is free; two research tools are available on request while we finish them.

Open-source tools

  • Open source

    Berry formerly Strawberry

    Checks whether an AI’s answer is supported by the evidence it cites. For each claim it measures how much the answer depends on that evidence, flags claims that don’t clear the bar, and keeps a tamper-evident record of every decision. Runs as a local tool inside Claude Code, Cursor, Codex and Gemini CLI.

  • Open source

    Mezzanine

    Measures how much a model’s predictions swing across symmetry-equivalent inputs, then distils a student model that averages the swing away. Comes with recipes for robotics, weather, particle physics and fast physics surrogates.

  • Open source

    ntkmirror

    Adapts Hugging Face language models with small controllers in the forward pass instead of LoRA. Controllers fitted separately can be combined by addition.

  • Open source

    triton-blackwell

    Makes Triton, a widely used tool for writing fast GPU programs, work on NVIDIA’s GB10 (Blackwell) desktop hardware until official support arrives.

  • Free to read

    Information Geometry for Generative Models

    Thirteen free chapters, from compression and Bayesian prediction to transformers and diffusion.

On request

Research tools we share directly while they’re being finished. Email us to try one.

  • On request

    Injection Audit

    Tests AI systems against whole families of prompt-injection attacks, varying wrappers, positions and encodings, and reports a statistical bound on how often they fail rather than a single pass or fail.

  • On request

    Optimal CoT

    Explains why longer chains of reasoning can make answers less reliable, and how to choose the right length for a task.

Our open-source work passed 2,000 GitHub stars in total; HallBayes, now Berry, passed 1,000 in its first month.