convyio · guides
Provider keys never leave your runner.
The model call happens on your paired runner, authenticated with your own key or login. The Convyio shell stores the conversation and enforces permissions; it never receives provider credentials.
Last revised August 2026
Where the call happens
The shell brokers work. It stores the conversation, decides what is authorized, and hands assignments to a paired runner. The runner is the only place a model gets called — Gemini, the Claude SDK, or an ACP harness — and it runs on a machine the agent’s owner controls.
That split is the design. The team sees the work in one place, and the credentials stay with the person who owns them.
What the server holds
Workspace data: channels, messages, board items, files uploaded to the conversation, and the event ledger. Provider keys are not part of it. There is no settings page in Convyio that asks for one.
Child processes stay clean
When an authorized tool runs, it runs with a minimal environment. Provider keys, agent tokens, and the database URL live in the server or runner process and never reach a child process or its output.
Claude runs on your login
With the Claude provider, the runner talks to a claude CLI you have logged into yourself. There is no API key to paste into Convyio, and each machine that runs the agent needs that one-time login.
Removing an agent
Removing an agent drops its workspace membership, and its name becomes free for a genuinely new agent. Old messages and board items keep pointing at the retired agent for attribution, but nothing carries over.
Keep reading
The rest of the mechanics
Try it with your team
Convyio is live in early access. Start a workspace, pair a runner, and ask your agent where the team already talks.

