Tools in service of a research question

Tools & Infrastructure

Make threat models inspectable, connect claims to evidence and test the assumptions that matter.

Our internal threat-modeling platform

RiskGraph

Define risk pathways, document evidence and counterarguments, compare judgments and examine uncertainty. RiskGraph is COAI’s internal working environment for this research.

Follow the screenshot story to see how an explicit model becomes a question for further research.

See how we use RiskGraph

RiskGraph tutorial showing a network of five defined events — An explicit model makes dependencies available for review. Inside COAI’s internal RiskGraph platform · illustrative example, not a research result.

Mechanisms & research infrastructure

Platforms and libraries that let researchers inspect models without owning the hardware. Open source and open access wherever possible.

Library

mlxterp: Mechanistic Interpretability on Apple Silicon

21 November 2025 · Open source · MIT

Interpretability tooling assumes you have an NVIDIA GPU. mlxterp brings activation capture and interventions to MLX on Apple Silicon, so a MacBook is enough to start.

eDIF: A European Deep Inference Fabric for Remote Interpretability of LLMs

14 August 2025 · Pilot deployment

Inspecting a 70B model's internals needs GPUs most European labs do not have. eDIF is a shared remote fabric that lets researchers run interpretability experiments without owning the cluster.

Why we build this

A useful threat model exposes its assumptions and the evidence behind them. RiskGraph supports that scrutiny at the level of risk pathways; interpretability tools can supply additional evidence when a model mechanism matters to a research question.

Shared infrastructure helps researchers inspect systems without owning the full hardware stack. We build tools that make the reasoning and experiments easier to examine, while keeping the strength and limits of the evidence visible.

See our research