Current stageR12-R20 local foundations
Live development scope

What is in motion now

updated from the public state
  • R4.1-R10Protocol + conformanceGuarded / conforming
  • R12Worker boundaries + permission receiptsImplemented locally
  • R13A2A / ACP transport + adaptersImplemented locally
  • R14Model Atlas + usage ledgerImplemented locally
  • R15Control rooms + private projectionsImplemented locally
  • R16Library intake + progressive onboardingImplemented locally
  • R18Trusted devices + recoveryFoundation / gated
  • R19Authenticator + mobile approvalsFoundation / gated
  • R20Multi-authority queues + routingFoundation / gated
  • M5Hosted tenancy + external operationsNot enabled

Green means the local contract or conformance work is implemented. Yellow marks a foundation whose production boundary is still gated. Red means the capability is intentionally not enabled.

Menu
A clearer starting point

What is Relay?

Relay is an evidence and governance layer for work that moves between AI tools, human reviewers, and teams.

See how it works Try the local path Read the FAQ
The short version

Many models can help. One accountable record keeps the work understandable.

Relay takes disconnected sessions, files, reviews, and hand-offs and turns them into a bounded task, attached evidence, visible disagreement, and a named human decision.

Who is it for?

Useful anywhere the answer matters more than the first draft.

01 · Solo builders

People using several AI tools

Keep transcripts, assumptions, evidence, and decisions together when you move between ChatGPT, Claude, Codex, Grok, Gemini, or local tools.

02 · Research teams

Work that needs reproducibility

Preserve methods, environments, failures, reviewer independence, and remediation instead of relying on a polished final paragraph.

03 · Engineering teams

Changes that need a trail

Connect tasks, proposals, reviews, pull requests, and human approval without making a model the merge authority.

04 · Operations

Knowledge that must survive hand-offs

Give Notion, Slack, GitHub, and local control rooms useful projections while keeping the canonical record rebuildable and private by default.

The Relay loop

Four moves from prompt to accountable outcome.

  1. 01Bound the question

    Declare the task, owner, risk, inputs, constraints, and acceptance criteria.

  2. 02Capture the work

    Normalize transcripts and attach methods, commands, artifacts, failures, and hashes.

  3. 03Compare and review

    Keep agreement and disagreement visible; require reproduction when the policy says it matters.

  4. 04Leave authority with a person

    A named human accepts, remediates, rejects, or defers the consequential decision.

Boundaries by design

What Relay is—and what it is not.

Relay is
  • Provider-neutral evidence and governance infrastructure
  • A local-first archive and comparison workflow
  • A rebuildable projection into control rooms
  • A record of provenance, review, disagreement, and authority
Relay is not
  • A model router that requires paid APIs
  • A truth engine or automated consensus oracle
  • A replacement for expert or scientific review
  • A system that lets models approve consequential decisions
Continue

Choose the depth you need.

01

How it works

Follow the evidence gate and canonical record flow.

Open protocol → 02

Control rooms

See how local, Notion, Slack, and GitHub surfaces fit.

View control rooms → 03

Get started

Run the public-safe local workflow without provider keys.

Download locally → 04

FAQ

Answers about privacy, models, decisions, and current limits.

Read answers →