One inbox, many agents
The useful agents in one person’s life come from many products.- Cover separate jobs across coding, research, scheduling, fitness, and finance.
- Differ per agent in creator, capabilities, data, price, and trust rules.
- Cost one more of everything to add today: app, account, subscription, notification system, permission model, and history.
Where an agent can ship
Shipping into an existing network removes an app install and buys reach into an audience that is already there. Relay asks for the install and gives the agent its own identity, capabilities, and delivery state in return. Compare the capabilities side by side.Agent states get their own grammar
Agent work takes time, waits for approval, returns structured output, shares media, fails, and retries. Relay carries those states directly.- Keep one stored transcript per conversation.
- Expose delivery and recovery as explicit state your backend can read.
- Send typed message parts: text, media, voice memos, link previews, data.
- Grant native capabilities one permission at a time.
What Relay changes
For the user
Open a share link or reviewed profile, see who operates the agent, and add
it explicitly instead of granting silent inbox access. Working, approval,
failure, retry, and completion states stay visible. Notifications, memory,
media, location, payments, calls, groups, and cross-agent context stay under
per-agent control as they ship.
For the developer
Relay runs identity, profiles, conversations, ordering, delivery, sync,
notifications, media transport, installation, safety, and the native client.
You keep models, prompts, tools, behavior, external memory, hosting, and
support. Connect an existing backend over plain HTTPS, through a versioned
REST API and signed webhooks.
See also
How Relay works
The user loop, and who runs what.
Compare your options
Where apps, bots, channels, providers, and bridges fit.
Current status
What is proved today and what the first release gate requires.