Prompt Architect
Understands the project, researches the work, maintains Project Brain context, and prepares the structured handoff.
CURRENT DEMONSTRATION / GUIDED MODE
Relay coordinates your Prompt Architect, Coding Agent, Reviewer, and project context in one workspace—without manually copying work between disconnected AI tools.
One project. Shared context. Controlled permissions. Verified work.
PRODUCT VISIONBUILD CUSTOM COMPOUND AI AGENTS↓
RELAY: Project context secured. The next handoff can continue without rebuilding the work from memory.
▼CURRENT PRODUCT / COORDINATED WORK
CURRENT DEMONSTRATIONEach participant receives a clear role, owned work, approved access, and the project context required to continue. Relay controls the route so humans and agents do not duplicate work or lose decisions between tools.
Understands the project, researches the work, maintains Project Brain context, and prepares the structured handoff.
Owns the assigned implementation, inspects the project, makes permitted changes, runs checks, and reports a claim.
Independently reviews the result, creates findings, requests repairs, and approves only when the evidence is sufficient.
Preserves context, controls ownership and permissions, separates claims from evidence, and determines when completion is earned.
COMPOUND AI AGENTS / PSP
IN DEVELOPMENTA Relay Compound AI Agent is an orchestrated system—not a single model.It combines a Prompt Architect, Coding Agent, Reviewer, Project Brain, approved tools, bounded sub-agents, permissions, and verification rules into one reusable workforce.
Depending on its connected models, tools, and Project Leader permissions, a compound agent can research, plan, inspect repositories, do everything expected of a Coding Agent, verify results, prepare deployments, and create specialized AI sub-agents for defined work.
The active, coordinated workforce that performs the mission.
The portable blueprint that reconstructs how that agent operates.
The planned publishing layer for free, open, source-visible, or paid agents.
LEAD RELAY DOGCOMPOUND CORE / READYPrompt Architect, Coding Agent, Reviewer, and Relay Core
SETDurable Project Brain context shared across the workforce
SETApproved repositories, commands, connectors, and services
SETTask-bounded specialist agents represented by cub puppies
SETPermissions, limits, approvals, and stop conditions
SETVerification rules and an explicit completion policy
SETHOW RELAY WORKS / ONE CONNECTED ROUTE
Relay keeps every handoff, decision, claim, finding, and repair connected to the same project.

Describe the project or task once.
Builds the brief, researches the project, and prepares the handoff.
Receives one owned task, implements it, runs checks, and reports the work.
Separates the agent’s claim from independently checked evidence.
Approves the result or creates findings and requests a repair.
MISSION PREVIEW / 001
01Inspect the route from owned assignment to returned evidence. The live demonstration remains the shortest path into Relay.

STEP 01 / TASK SHAPED
The Architect Twin turns the goal into scoped instructions, constraints, acceptance checks, and the exact context the next agent needs.
PROJECT BRAIN / DURABLE CONTEXT
Project Brain gives every approved human and connected agent the context it needs. Change the Coding Agent, invite a teammate, or return days later without rebuilding the project from memory.
Project Brain belongs to the project—not to one AI provider.RECENT HANDOFF / BRIEF_018 · 12 CONTEXT ITEMS · READY
RELAY MULTIPLAYER
COMING SOONInvite your team into Relay, assign clear roles, and control what every person and connected agent can see, change, approve, or execute.
Owns the project, controls permissions, connectors, limits, invitations, and sensitive approvals.
Maintains the Project Brief, settings, tasks, and shared Project Brain when permitted.
Works on assigned implementation tasks and submits work for verification.
Creates findings, requests repairs, and approves or rejects assigned changes.
Observes approved project history or receives narrow, temporary access.
PSP AGENTS / PORTABLE BLUEPRINT
IN DEVELOPMENTA Project Settings Preset (PSP) does not save one prompt. It is the portable blueprint for a compound AI agent: its models, roles, routing, tools, Project Brain, cub sub-agents, permissions, limits, verification rules, completion policy, and Relay Dog identity.
Publish source-visible configuration under a chosen license so others can inspect, install, and—when allowed—fork it.
Set paid access, license terms, update rights, and the parts of the agent configuration buyers may inspect.

RESEARCH CUB
VERIFY CUB
BUILDER CUBRELAY DOG PACK / AGENT IDENTITY
PRODUCT VISIONThe lead Relay Dog represents the Compound AI Agent. Cub puppies represent the specialist sub-agents it creates or assigns. Each cub receives one bounded task, approved tools, and limited permissions, then reports its evidence back through Relay.
Finds evidence and returns cited context.
Runs bounded checks and reports evidence.
Creates or configures a specialist agent.
Customize the pack's name, colors, eyes, collars, badges, accessories, and animation style. The identity remains attached to its PSP.
SHIP ON SUNDAYS / SOS
PRODUCT VISIONShip on Sundays is the planned marketplace for PSP agents. Creators will be able to publish a free open-source or source-visible agent, or offer a commercial agent through paid access with clear license, update, permission, and compatibility terms.
PSP DOG / ORBITAlcatraz planning, Claude Code execution, three specialist cub agents, Guided approval, Project Brain, and independent verification.
Before activation, the Project Leader reviews:
Publish the PSP configuration with a clear open-source or source-visible license. Define whether others may inspect, fork, modify, and redistribute the agent.
FREE · LICENSED · FORK RULESPlace paid access around a compound agent, disclose what buyers receive, and define project, team, update, and source-access rights.
PAID · LICENSED · VERSIONEDEvery install still shows requested tools, files, commands, network access, limits, verification, storage behavior, and Project Leader approvals.
NO AUTOMATIC PERMISSIONSACCOUNTABLE MISSION HISTORY
CORE PRODUCT PRINCIPLERelay records who did the work, what was claimed, what was verified, what was repaired, and who approved the final result.
The original objective and approved Project Brief.
Who accepted the task and what they were permitted to change.
What the Coding Agent says it completed.
Checks, file changes, Relay verification, and reviewer findings.
What was returned, repaired, and independently re-reviewed.
Which policy passed and which human approved the final result.
CONTROL DEPTH / OPERATING MODES
Every mode remains permission-bound and observable. Autonomous never means unrestricted.

Relay handles routine coordination and asks for human approval at important transitions.
CONTROLLED · OBSERVABLE · FOUNDER DEMO▶Relay continues through approved categories of work and pauses for higher-risk or ambiguous actions.
SELECTIVE CHECKPOINTS · BOUNDED SCOPE+Relay operates inside predefined permissions, limits, completion rules, and visible stop conditions.
OBSERVABLE · LIMITED · NEVER UNRESTRICTED+PRIVATE BETA / CURRENT DEMONSTRATION
Start with the current Guided workflow. Join the product vision for custom Compound AI Agents, PSP publishing, Relay Dog cub sub-agents, Multiplayer, and Ship on Sundays.
OPEN RELAY↗JOIN THE RELAY WAITLIST ↗Built by Aquala Technologies · Part of the Sunday family