One library, not forty tabs.
Sources, files, brand knowledge, and prompt templates your admins actually approved — reusable by anyone on the team.
Human + AI task operations
Relay turns a request into a deliverable — routed through your approved prompts, your context, your agents, and a person when judgment matters. One library, one meter, one place it all runs.
Free to start. Usage billed in $25 credit blocks — no card until you run paid work.
From an AI free-for-all to running it like the best companies in the world — without the seven-figure implementation, the dedicated team, or the training programme.
02 · The problem, named
Different tools, different accounts, prompts pasted into notes apps, no shared context, no record of what was made or what it cost. It works until you need it twice.
03 · What Relay is
Sources, files, brand knowledge, and prompt templates your admins actually approved — reusable by anyone on the team.
A single AI step, a repeatable workflow, an agent that adapts, or a person who signs off. You choose per task; Relay remembers what worked.
Docs, tables, timelines, decks — plus the sources, model, cost, and approvals behind them.
04 · Two ways in
For companies
Invite your team, publish approved prompts and workflows, meter spend centrally, connect your systems once. Members never see a bill.
For individuals
Keep your context, prompts, workflows, and agents in one place — and rent proven ones from creators instead of starting cold.
05 · Marketplace
Publish a prompt, workflow, or agent. Draft and test it privately, set a price per output, connect Stripe, get paid every time someone runs it. Buyers get something already proven; you get the margin on work you have already done.
06 · Governance, plainly
See what every model call costs and who made it. Restrict which models are allowed. Bring your own keys when you are ready. Keep approval on the work that carries risk. This is the part that usually costs seven figures and a dedicated team.
Work receipt
Market research brief · run 0248
07 · Bring one job
Run it free, keep the output, and decide from there.