Frequently asked questions

01What exactly is Relay?+

Relay is a provider-neutral coordination and evidence layer. It records bounded tasks, model or human contributions, evidence, reviews, disagreements, and the final human decision in a durable, inspectable history.

02Who needs Relay?+

Solo builders using multiple AI tools, research teams, engineers reviewing AI-assisted changes, and operations teams managing knowledge across chats and control rooms. If you only need one disposable answer, Relay may be more than you need.

03Does Relay replace ChatGPT, Claude, Grok, Codex, or other models?+

No. Those tools remain execution surfaces. Relay gives their work a shared intake and review contract without requiring one vendor or one orchestration API.

04Do I need paid APIs or a universal model gateway?+

No. The free reference path is local: CLI, Library, manual packets, and optional local models. Provider APIs and hosted services are opt-in accelerators, not requirements.

05How do models communicate through Relay?+

They can contribute through local transcript ingestion, MCP, GitHub proposals, manual packets, or guarded projections. Relay normalizes and compares what was submitted; it does not pretend that different model brands automatically prove independence.

06Can Relay decide which answer is correct?+

No. The Overseer can identify agreement, conflict, missing evidence, and suggested routes. Only the named human authority can accept, remediate, reject, or defer a consequential decision.

07Where is my archive stored?+

The local archive is local-first and ignored by Git. You can keep it on disk, use encrypted backups, or explicitly choose a private storage projection. Private transcripts stay local by default. If you enable the Git archive adapter, Relay requires an explicitly selected private repository and an approved encrypted bundle; Pages and public exports never receive the archive.

08What are Notion and Slack used for?+

They are control-room projections for knowledge, notifications, and hand-offs. They are not canonical authority, and guarded connectors cannot write a human decision on a model's behalf.

09What is available today?+

R4 is shipped, and the local R12–R20 foundation is implemented as tested contract boundaries: restricted framework adapters, A2A/ACP transport, Model Atlas evidence telemetry, private projections, guided onboarding, Jira, trusted devices, mobile approvals, and multi-authority queues. These local boundaries remain non-authoritative and provider-neutral.

10What is still being built?+

Production custody and key recovery, real remote peers and Jira tenants, native authenticator/mobile operations, authenticated external sinks, independent reviews, live provider measurement, and the optional hosted tier remain guarded, incomplete, or deferred. See the roadmap for the current sequence.

11Is Relay a truth engine or scientific peer review replacement?+

No. Schemas and hashes prove structure and integrity, not truth. Relay preserves the evidence and review conditions so people can make better-informed decisions.

12Can I ask several local models the same question?+

Yes. The desktop Review tab includes a guarded round-table chat. Select the workers you have installed, send a bounded question, and inspect each reply plus the deterministic comparison. The result is advisory and only a human can record the final decision.

13How do I try it?+

Start with the local installer or clone the repository, run npm install, then npm run check. The Download page explains the public-safe path.