# RunVect AI — Revon

> RunVect AI is a self-correcting Agent runtime. Revon helps people and AI Agents turn owner-confirmed context into proposed actions, evidence-backed outcomes, and review-gated learning while the owner keeps final authority.

## Public product surfaces

- [RunVect AI](https://runvectai.com/): Product overview and access to Revon.
- [Agent Access](https://runvectai.com/agent-access): Public description of the External Agent contract, approval boundary, and available operations.

## Machine-readable Agent contract

- [External Agent Early Access Manifest](https://runvectai.com/api/agent/early-access/manifest): Live public JSON contract for supported operations, authentication, scopes, and owner-approval requirements.
- [Public Agent Developer Reference](https://runvectai.com/agent-developer-reference.md): Generated endpoint reference derived from the OpenAPI contract.

## Runtime principles

- External writes require explicit owner approval through an Action Run.
- Agent credentials are owner-bound, scope-limited, expiring, and revalidated at use time.
- Outcomes and evidence remain distinct from claims; learning candidates require owner review.
- Supported Action Run integrations include GitHub issues, Google Calendar events, Microsoft Outlook Calendar events, and Slack messages.

## Authority boundaries

- Public discovery does not grant access to private records or external services.
- Protected operations require an Owner-issued Bearer credential with the exact declared scope.
- Provider OAuth access never substitutes for owner identity, delegation, or Action Run approval.
- Do not infer success, trust, or learning from a proposal alone; inspect the returned Action Run state and evidence.