Relay HQ

CHALLENGE

A real operational problem, not a demo.

Growth work for a mobile app is spread across analytics dashboards, store listings, competitor research, content calendars, ad budgets, and support inboxes. Most of it is repeatable, but none of it talks to the rest — and the team did not want an AI with free rein over live accounts.

SOLUTION

A system built around the work.

Relay HQ is the platform layer: auth and roles, isolated workspaces, a product registry, and an agent-run executor wrapped around the Claude Agent SDK. Growth Agent is the first product on top. Users only talk to Morgan, the Growth Lead. Morgan plans the work, delegates to specialists in parallel where it can, waits on dependencies where it must, and merges findings into one recommendation.

THE TEAM

One lead. Six specialists.

Morgan
Growth Lead
Market Research
Competitors, ASO, audience
Analytics
Funnels, retention, anomalies
Content Growth
Ideas, hooks, post drafts
Paid Ads
Budgets and forecasts
Customer Ops
Reviews and support themes
Partnerships
Distribution opportunities

WHAT MAKES THIS DIFFERENT

The automation stays accountable.

Every workspace resolves through a single context chokepoint, so one company's data never leaks into another's. Runs queue through Inngest with retries and token accounting. The Knowledge Hub holds brand, competitors, and research — agents can only propose changes; a human approves them.

HOW IT WORKS

A clear path from signal to action.

Ask Morgan

A team member asks a question or sets a goal in the chat.

Plan and delegate

Morgan builds a dependency graph and hands work to the specialists that fit.

Run in the background

Inngest executes each run with retries; progress streams back as a replayable event log.

Merge and recommend

Morgan reconciles conflicting findings and returns one recommendation with follow-up tasks.

RESULTS

From fragmented work to a plan people can use.

Phases 0–5 are built: platform shell, office, agent runtime, Knowledge Hub, the specialist team, and Drafts. Automated tests, typecheck, and lint stay green. Integrations and instruction authoring are next — the operating surface is already real.