deco Studio

Built-in Tools

Core tools provided by decopilot for discovering capabilities and managing work

Overview

Decopilot provides built-in tools that are always available (subject to task/subtask restrictions). These tools enable work management, context access, and agent coordination.

Tool Categories

Discovery & Enablement

  • enable_tool — Activate one or more scope tools so they can be called in subsequent steps

Task Management

  • subtask — Spawn a subtask with isolated context, optionally bound to a specific agent

User Interaction

  • user_ask — Ask the user a question when input or clarification is needed

Context Access

  • read_resource — Read documentation or guidelines from the current scope
  • read_prompt — Read a reusable prompt template

Working with Tool Output

  • read_tool_output — Re-read a large tool result that was offloaded to storage

Research and Media

These tools are registered conditionally based on the organization’s configured default models:

  • web_search — Stream a deep web-research pass (requires a web-research model)
  • generate_image — Generate an image (requires an image model)

Planning

  • propose_plan — Propose a plan for approval. Only registered when chat mode is plan .

Sandbox (VM) tools

When an agent is backed by a sandbox (a linked GitHub repo), decopilot also exposes six VM tools — bash , read , write , edit , grep , glob — that operate inside that sandbox. They’re provisioned lazily on first use.

Availability Matrix

Tool Task Subtask Notes
enable_tool Activates a scope tool for the next step
subtask Prevents infinite delegation
user_ask Subtasks can’t interrupt the user
read_resource
read_prompt
read_tool_output
web_search Conditional on web-research model
generate_image Conditional on image model
propose_plan Plan mode only
VM tools ( bash , read , write , edit , grep , glob ) Only when agent has a sandbox

Tasks are user-initiated and have full access to all built-in tools. Subtasks are spawned by decopilot and have restricted access to prevent infinite delegation and blocking user interactions. Subtasks have a step limit of 15 (vs 30 for tasks) and must return a summary.

Tool discovery: Available scope tools from connected MCP servers are discovered implicitly via the connections listed in the current scope’s Virtual MCP — there is no explicit search tool. Use enable_tool to activate a tool before calling it.

Found an error or want to improve this page?

Edit this page