One accountable record
Tasks, events, evidence, reviews, and decisions have stable identifiers, deterministic hashes, and causal history.
Evidence before consensus
One record for work
across many minds.
Relay turns disconnected AI sessions and human reviews into bounded tasks, reproducible evidence, explicit disagreements, and accountable decisions.
Why Relay exists
AI can produce more answers than one person can responsibly inspect.
Relay is the coordination layer between AI tools, reviewers, and the person accountable for the result. It keeps the question, supporting evidence, disagreement, and final human decision connected—even when the work crosses models or applications.
The rule
More agreement is not the same as more truth. Relay makes the path to a decision inspectable.
What works today
A measured system, not a promise.
Evidence without execution
Methods, commands, environments, artifacts, failures, and hashes can travel with a claim without Relay executing them.
Disagreement stays visible
Independent review and reproduction are explicit. Contradictions trigger remediation; they are not silently averaged away.
Humans keep authority
Models can contribute and compare. A named person still accepts, remediates, rejects, or defers consequential work.
One protocol, many contexts
Change the evidence threshold, not the record.
Multi-model work
Compare competing outputs without losing who said what, what they used, or where they differ.
Operations
Bound incidents and changes with reproducible checks, visible exceptions, and accountable decisions.
Research
Preserve methods, artifacts, independent reproduction, adversarial review, and named scientific authority.
Governance
Keep evidence gaps, residual risk, review trails, and final approvals legible to the people responsible.
Project control plane
Roadmap and blockers
Loading status?
Boundary-safe proving grounds
Pilot readiness
Release route
Ship the safe path first.
Foundation
Protocol kernel, evidence gates, local MCP, and public-boundary checks.
completePublic-safe foundation
GitHub contract, Knowledge Hub projections, and read-only discovery CLI.
shippedOverseer MVP
Normalize transcripts, compare model intakes, and queue human decisions.
next gateControl rooms and scale
Notion, Slack, live archive, MCP setup, routing, and optional hosted services.
plannedControl rooms
One governed record, different ways to work.
Local-first
CLI, local archive, and optional local model: the free portable reference path.
R7–R9Notion
A structured knowledge and task projection: convenient, rebuildable, never canonical authority.
R5Slack
Threads, notifications, review requests, and attributed commands. Messages never equal approval.
R6GitHub
Reviewable proposals, issues, receipts, and public-safe operational status.
R1–R4Build map
Every convenience layer has an evidence gate.
Foundation
Canonical records, policy gates, local MCP, public-boundary checks.
CompleteSafe collaboration
GitHub contract, Knowledge Hub projections, and read-only discovery CLI.
CompleteOverseer MVP
Transcript normalization, comparison packets, assignment, provenance, redaction, and human approval queue.
Next gateControl rooms
Opt-in Notion projection, then Slack channels, threads, notifications, and signed commands.
PlannedArchive and installer
Low-RAM local bookshelf, retention/export, MCP discovery, dry-run configuration plans, backup and rollback.
PlannedRouting and scale
Provider adapters, manual packets, local-model fallback, cost policy, then optional managed hosting.
FutureCanonical projection
Current tasks
Loading state…
Task
State
Risk
Evidence
Reviews
Default evidence gate
From question to decision
- 01Bound the taskOne question, inputs, constraints, and acceptance criteria.
- 02Submit evidenceMethods, commands, environment, artifacts, failures, and hashes.
- 03Review independentlyReproduce where required and preserve material disagreement.
- 04Record authorityA named human accepts, remediates, rejects, or defers.