Read World-State
Agents may retrieve compact state and timeline artifacts.
Agent safety
Agents can read evidence; they should not autonomously make destructive real-world decisions.
Ayneye supports evidence-first agent workflows. In the controlled beta, agent-facing behavior must be read-only by default, budgeted, tenant-scoped, and reviewable.
Allowed by default
The safe path starts with reading and summarizing evidence.
Search evidence
Agents may search evidence bundles and timestamps.
Ask with evidence
Answers should include evidence references and uncertainty.
Respect budgets
Tool calls and executions count against plan limits.
Not allowed by default
These actions require explicit future approvals and separate governance.
Unlocking doors
No physical access control automation from beta video understanding.
Dispatching emergency response
No automated high-stakes action without human review.
Identity/intent conclusions
No unsupported claims about identity, intent, or emotion.
Destructive API calls
No write/destructive agent tools in the default beta.