Agent Studio Overview
Agent Studio puts Claude Code inside an environment built for shipping Sapiom agents. You keep the coding-agent conversation and terminal while Studio adds project navigation, deterministic inspection, and direct local or cloud lifecycle actions around it.
It’s your coding agent on your machine: your login, plan, and model settings apply, and sessions run locally. Cloud operations such as capability calls, deploys, and production agent runs go through Sapiom.
The layout
Section titled “The layout”Three surfaces, left to right:
- Workspace rail. Open local folders, discover agent projects, start from a template or idea, and choose the project a session targets. See Workspaces and projects.
- Sessions. Run parallel Claude Code conversations in real terminal processes, review recorded history, and choose native resume or a new conversation with a bounded brief. See Sessions.
- Canvas · Steps · Code. Read the bound agent’s deterministic graph, inspect authored steps and observed runs, then get SDK or HTTP starters after a cloud build is ready. See Canvas, Steps, and Code.
Press ⌘K to jump between sessions, agent projects, and recent folders, or to browse templates. Local Run, Prod Run, and Deploy stay in the action bar above the terminal.
Two kinds of action
Section titled “Two kinds of action”Studio keeps direct product actions separate from coding-agent requests:
- Direct actions run through Studio itself. Canvas rendering and Local Run are local; Deploy and Prod Run call Sapiom cloud. None prompts Claude.
- Coding-agent actions type or inject a request into Claude Code. Free-form chat, step debugging and explanation, AI-authored descriptions, and template or idea creation belong to this path.
Local inspection and local runs can work signed out. Cloud deploys and production runs require a Sapiom account; production work can incur capability spend.
Session-scoped wiring
Section titled “Session-scoped wiring”Studio launches Claude with session-scoped authoring instructions and the local sapiom-dev MCP tools. This wiring lets Claude scaffold, check, locally test, deploy, and run agents without changing your global Claude Code configuration. Studio’s own local state and session records are separate and can remain after the terminal process exits.
© 2026 Sapiom, Inc.