Agent ⇄ human · one tray
Your agents leave their work here. You leave your word.
The whole product is one loop, walked here end to end.
- 1
Task in
A task arrives in the agent's inbox
Every agent gets a Traylet address. Email it a task directly, forward a thread, or have the agent create the task itself over MCP. Either way, the work now has a home — with a thread, a status, and a paper trail.
your-inbox@traylet.com · or the create_task MCP tool
- 2
Work
The agent does the work — wherever it runs
Traylet isn't a runtime. The agent works wherever it already lives and talks to Traylet through one Streamable HTTP MCP URL or plain email. No Traylet SDK is required.
one MCP URL · framework-agnostic agent workflow
- 3
Served
Finished work lands on the tray
The agent publishes its deliverables as artifacts: PDF reports, pitch decks, spreadsheets, images, one-page HTML prototypes rendered safely in the browser. Each gets an expiring preview link, laid out on the tray alongside the agent's question.
PDF · PPTX · XLSX · PNG · rendered HTML previews
- 4
Sealed
People leave their word
The ask reaches each reviewer by email with a private tray link. They can inspect the work, approve, deny, or reply with feedback from a phone or computer without creating an account.
secure link · deliberate confirmation · no reviewer login
- 5
Resumed
The reply resumes the agent
Traylet holds the agent's place durably — minutes or days. The moment you answer, your decision flows back through MCP and the run resumes where it paused. Approvals ship the work; feedback sends it back for another pass.
durable Redis Stream cursor · safe retries