01
An agent inventory
Register and inspect agent profiles with an identifier, name and operational status.
AFFINITY / Agent Trust
Agents that act.
Limits that matter.
Register your AI agents, limit their authority and evaluate actions before execution. A control layer alongside your orchestrator and business systems.
Pilot integration · scope agreed with you
Agent TrustInteractive example
Every action has a limit.
Support agent
Runningagent_demo_01Choose an example action
Allowed
Within delegation
ACTION_ALLOWEDTry the controls · simulated data
01 / The problem
An agent needs permissions, not unlimited freedom.
Reading a file is not exporting it. Preparing an operation is not executing it. When agents start acting, a broadly scoped API key leaves too many decisions implicit.
02 / What's included
Identify, delegate and control.
02
Task-based authority
Model local, expiring and revocable delegations together with action capabilities.
03
Runtime authorisation
Evaluate action requests and receive a decision with technical reasons. Your integration enforces the decision.
04
A stopping point
Activate a persistent kill switch checked during local authorisation. External propagation must be validated.
03 / How it works
From a registered agent to a bounded action.
- 01
Register the agent
Create an agent profile linked to your organisation and its operating context.
- 02
Limit its authority
Configure local delegation and the action capabilities you need to test in the pilot.
- 03
Evaluate and control
Check authorisation before acting and enforce the result in the target system. The kill switch blocks new local authorisations.
04 / IN YOUR CONTEXT
Built around real problems.
Every integration starts with a specific use case. Here are a few starting points.
Let's talk01Support with boundaries
Test an agent that can read tickets but cannot export data or change customer identity. Define concrete actions and enforce decisions in the support system.
02Assisted operations
Allow information preparation or reconciliation while keeping sensitive execution under human and target-system control.
03Professional assistants
Limit an assistant to preparing drafts or gathering information. Signing, filing and final decisions remain outside its delegated authority.
PRODUCT SCOPE
Clarity from day one.
What you can explore today
Pilot registration, local delegation, per-action authorisation and kill switch. Evidence generation is attempted and is not guaranteed for each decision.
Where it is heading
End-to-end verifiable delegation, consent, unified receipts and propagation guarantees remain in development. The product does not execute agents or replace their orchestrator.
Agent Trust
Your use case. Our next conversation.
Tell us what you want to solve. We’ll define the scope and the next step together.
Let's talkView overview