Overview - Deco CMS

What is decopilot?

Decopilot is the AI assistant built into deco Studio. It can work across all your connected tools — Shopify, databases, email, customer service platforms, and more — from a single chat.

You start a conversation and decopilot works in the background. When it needs your input, it asks. Otherwise it keeps going, coordinating across multiple services on your behalf.

How it works

Decopilot follows a simple loop: understand the task → take action → check the result → repeat. It reads from your connected services, calls the right tools, and adapts based on what it learns.

What makes it different from a regular AI chat is that it can reach into the apps where your actual work lives — it’s not limited to answering questions. It can look up orders, update records, send notifications, and coordinate workflows across multiple services in one go.

Scope: where you’re working determines what’s available

Decopilot adapts to where you are in Studio:

Organization scope — you’re at the home screen with no agent open. Decopilot has access to org-wide built-in tools (managing connections, agents, members) and your full agent catalog.
Agent scope — you’ve opened a specific agent. Decopilot now uses that agent’s tools and instructions, with the agent’s thread history as context.

To switch, just open or close an agent from the sidebar. There’s nothing extra to configure.

A third level — Project scope — is planned. It will sit between org and agent and group tools/context for a team or client. See Projects.

Chat modes

The chat input has a mode selector that shapes how decopilot approaches a turn. You can switch modes at any time without losing your conversation.

Mode	What changes
Default	Normal chat — decopilot picks tools on its own.
Plan	Decopilot must end the turn with a `propose_plan` call. Destructive tools are blocked until you approve a plan; only read-only tools can be enabled. Useful before doing something risky.
Web search	The first step is forced to call `web_search` , which streams a deep research pass directly to you. Requires a web-research model under AI Providers.
Gen image	The first step is forced to call `generate_image` . Requires an image model.

Models

Decopilot works with multiple AI providers — Anthropic, Google, OpenRouter, the Deco AI Gateway, and self-hosted models. Pick a stronger model for complex tasks and a faster one for simple operations. See AI Providers.

Agents and subtasks

When a task needs focused work — a customer-support question, a research pass, a data analysis — decopilot can hand it off to a specialist agent in a subtask. The subtask runs independently with its own clean context, then reports a summary back. Your main conversation stays clean.

Built-in tools

Decopilot ships with tools for coordinating work across any scope:

Tool	What it does
`subtask`	Delegates work to a specialist agent
`user_ask`	Asks you a question when it needs clarification
`enable_tool`	Activates a scope tool for use in subsequent steps
`read_resource`	Reads documentation or guidelines available in scope
`read_prompt`	Reads a reusable prompt template
`read_tool_output`	Re-reads a previous tool result that was offloaded to storage
`web_search`	Runs a deep web research pass and streams the result
`generate_image`	Creates an image with the configured image model
`propose_plan`	Proposes a plan for your approval (plan mode only)

When an agent is backed by a sandbox (a linked GitHub repo), decopilot also gets six VM tools — bash , read , write , edit , grep , glob — so it can execute and modify code in that sandbox.

Sandboxes also mount the organization filesystem under org/ : a folder named after your organization (e.g. org/acme/ ) is the org’s shared home — editable, free-form, shared across every member, agent, and run, where agents record durable knowledge and check for context before starting work; org/upload holds the files attached to the current conversation (chat attachments land there automatically — no copy step); org/output is the current run’s shared output folder; and org/public/<set> exposes curated read-only skill sets synced from versioned repositories. Files written to the home folder and org/output sync to your organization’s cloud storage and are visible to every member and agent; external changes appear inside the sandbox within about a second.

macOS desktop links: when a sandbox runs on your machine ( deco link ), the org/ folders are network volumes, and macOS asks once per app for permission to access them. Approve the “access files on a network volume” prompt for your terminal or editor — or enable it manually under System Settings → Privacy & Security → Files & Folders → your app → Network Volumes. Without the grant, reads and writes in org/ fail with “Operation not permitted”.

For the full reference, see Tools.

Learn more

Context and tasks — how memory works, and how to use subtasks for heavy work

Tools — full reference for built-in and scope tools