AI Capabilities
AI Capability Guide
What's live, what's degraded, and what's coming next in the AI catalog.
Live and guarded
- AI execution routes require tenant auth, Zod payload validation, and prompt-injection checks.
- Runtime provider/model overrides are restricted to users with settings edit access.
- AI run history and feedback are visible through the platform surfaces already wired for the milestone.
Coming or limited
- Some roadmap capabilities remain visible-only until traceability marks them implemented.
- Dispute AI still needs seeded staging execute/feedback proof before it is fully closed.
- Production AI depends on a real configured provider and live outage/degraded-mode evidence.