Agent Studio
One agent. One job. One conversation.
An agent is a single reasoning unit you talk to — with its own prompt, tools, knowledge, memory, and revision history. Build it on the canvas in Studio, chat with it in the lab, and deploy it to the channels your users already use.
Let your coding agent build it
The fruxon CLI is self-describing: fruxon describe emits the entire command tree as JSON, and every write command hands back the schema for its own request body. Point Claude Code or Cursor at it and the agent looks up what each task needs instead of guessing field names. Drafts run tagged as tests, so iteration never lands in production metrics or budgets.
Placeholders, not string-stuffing
Pull values into a prompt by reference — {{param.topic}} for a typed input, {{secret.STRIPE_KEY}} for an org secret, or an attached asset from the editor's picker — and Fruxon resolves them at run time. You insert them from a menu instead of hand-typing, so a name never silently fails to resolve.
Production and Dev, side by side
Studio shows each agent two ways. Production is the deployed revision, its live executions, and its real spend. Dev is your drafts, test runs, sandbox-backed chat, and test-origin spend. You iterate on a candidate without touching the revision serving traffic.
How it fits together
An agent is the unit everything else composes.
Because an agent stays single-turn, it can be bound to a Slack, Teams, or Telegram channel and answer as a participant. It can be embedded in another agent as a sub-agent tool. And it's the unit a Workflow references when a job needs more than one turn of reasoning.
Every save captured as an immutable revision — compare, deploy, or roll back to any of them
Deployable to channels as a network participant, which multi-step definitions can't be
Callable as a sub-agent tool from another agent, or as a referenced step inside a Workflow
The rest of the build surface
Every capability is included on every plan — see pricing