Crucibleby Humanis
Book a demo

Verified AI for
critical infrastructure

Crucible models your system, proposes the actions that improve it, and verifies each one stays inside your limits before anything runs.

Book a demo
A landscape of flowing mathematical contours.
The Crucible app showing a world model of a mobile network in the Gangnam district of Seoul: a 3D city map with cell coverage, beside panels for coverage, capacity and the agent's decision rationale.
A Crucible session. The operator asks for congestion relief; the agent reports its simulated trials in a table and states that no change was applied.

Deploying on live systems with leading network equipment providers and mobile operators.

Working with
  • ARIA, the Advanced Research and Invention Agency
The technology

The technology

Crucible is a verification-centric AI agent platform providing safety and explainability. At its core sit world models: live, low-latency digital twins of your system. Agents interact with these models to plan, and we use them to prevent unsafe actions from being deployed.

World models

Model your system

A low-latency simulation of your system built for “what if” lookahead. Agents can use this to plan in a sandbox before touching the real world.

Shields

Block unsafe actions

Every action of the agent is rolled out many times under different conditions before being deployed in the real world. Actions that could take your system outside its operational design envelope, the limits it must hold as conditions change, are flagged and blocked.

Formal verification

Prove performance

Where possible, we are working towards formal verification: mathematically proving claims about the system. Early results include proving optimal control policies on small fabs.

Explainability

Monitor reasoning

Planning inside the world model builds a full audit log of what an agent’s beliefs were and why it decided to pick a given action at a given time.

The research behind Crucible
Solutions

Explore Crucible
in your domain

Another system? If you wish to deploy AI in a safe way in the real world, tell us about it.

Towards L5 network autonomy

We are deploying agents and world models across high-value networks, providing use cases including network slicing, energy optimisation, or rApp conflict resolution, all within each network’s operational design envelope.

How an engagement runs

How an engagement runs

  1. Quantify

    We analyse a time-bound, anonymised set of data from your system to show the gains that Crucible can deliver.

  2. Demonstrate

    We deploy a model on a small-scale system and measure performance.

  3. Roll out

    We extend the deployment across your whole system.

Book a demo

See it on your system

Start with an assessment that quantifies the improvements Crucible can deliver.

Book a demo