Buildonthecontrolplane
Everything you need to integrate with the platform, define ontology schemas, register tools and MCP servers, and run agents under a mandate.
Whatyoucanbuildagainst
Platform API
REST endpoints for ontology objects, agent definitions, policy, approvals and evidence records.
Authentication
OIDC federation for humans; issued non-human identities with short-lived credentials for agents and services.
Ontology schemas
Define object types, relationships, events, declared actions, rules and metrics as versioned artefacts.
Agent runtime
Register an agent with a mandate, run it with durable execution, and receive its evidence record.
MCP integration
Register Model Context Protocol servers, pin versions, declare operations and scope them per mandate.
Tool registration
Declare operations with argument constraints so the broker can authorise and perform calls on an agent's behalf.
Model registry
Register models with permitted uses and evaluation results, and express routing policy.
Evaluation harness
Define task-specific suites and run them as lifecycle gates in your own pipelines.
Webhooks and events
Subscribe to approvals, denials, policy decisions and lifecycle transitions.
Theoneconcepttounderstandfirst
Your code does not hold credentials for target systems, and it does not decide whether an action is permitted. It declares what it wants to do against typed objects, and the control plane decides.
Concretely: instead of calling an ERP API with a service account token, you register the operation you need, and an agent under a mandate requests it. The broker authorises against the mandate, performs the call with its own scoped identity, and records the invocation against the decision. Your integration code gets simpler; the governance becomes real.
The same inversion applies to models. You do not name a provider. You declare the requirement — capability, data sensitivity class, latency tolerance — and policy resolves it. This is what makes swapping a model a configuration change.
Quickstartshape
The order that gets you to a working governed agent fastest.
API documentation and SDK access are provided to customers and to evaluation partners rather than published openly, because the surface is still moving and we would rather you build against something current. Request access and you will get the current reference and a route to an engineer.
Whatispublictoday
Open repositories
Research artefacts, evaluation tooling and reference material published on GitHub.
Published models and datasets
Labs releases on Hugging Face where they are the right thing to release.
Architecture references
Reference architecture and the request path, published in outline in the trust centre.
Ontology patterns
Sector ontology patterns described on the solutions pages.
Questions
- Is there a free tier?
- No. This is enterprise infrastructure that requires an identity integration and a deployment decision, and a self-service tier would misrepresent what it takes to run it. Evaluation access is available.
- Can I use my own agent framework?
- Yes, provided it routes data access, tool calls and model calls through the control plane. That is what makes the mandate enforceable — an agent that reaches systems directly cannot be governed by anything.
- Which languages are supported?
- The API is HTTP, so any language works. SDK coverage is prioritised by what customers are building in; ask for the current position.
Bring us one operational problem.
You do not need a finished brief. Bring the problem — we will work out the next step together.
Or book a call with the team