← Capability Map
AI Agents
BetaImplementedai-agents
Why this capability exists
Because the volume of scattered property evidence exceeds what a person can hold, and the reading of it must still be attributable.
How it works
Specialised agents that extract, cross-check and route. Each has a declared remit and a declared limit: what it may conclude, and what it must hand to a human.
What can I inspect?
Agent collaboration
/inside-the-machine#agents →
Agents
/agents →
Boundaries
/boundaries →
Trust Runtime
/trust-runtime →
Depends on
Depended on by
Constitutional owner
No agent conclusion is authority
Constitutional status
No agent conclusion is presented as authority.
Architecture layer
Reasoning engine — Turns what is known into an ordered investigation with a recorded path.
Ecosystem reach
Related Simulator modules
None
Related agents
- Evidence ExtractorLimit: May not interpret what an extracted claim means for the decision.
- Reliability ClassifierLimit: May not raise confidence above the strongest source supporting the claim.
- Question RouterLimit: May not close a gap it routed; only the professional can.
- Dossier ComposerLimit: May not state anything the graph does not already support.
- Resume AgentLimit: May not silently discard an open question to make the state look complete.