# Tools & Capabilities

## Review Tool Access

1. Open **Settings → Permissions**.
2. Review the tool groups available to the agent.
3. Keep Safe Mode enabled for commands that can change or delete data.
4. Inspect an approval request before allowing the action.

![Approval queue with pending agent actions](https://neotask-marketing-assets-417007889150.s3.us-east-1.amazonaws.com/landing/agents/2026-08-12-r1/approvals-main-1280.webp)

## Overview

Tools are the actions your agents can take during conversations. When an agent needs to browse the web, run a command, read a file, or control a device, it uses tools.

Neotask provides a rich set of built-in tools covering file management, code execution, browser automation, device control, messaging, web search, and more.

## File Operations

Agents can read, write, and edit files in their workspace:

- **Read**, Read file contents
- **Write**, Create or overwrite files
- **Edit**, Make targeted edits to specific parts of a file
- **Apply Patch**, Apply multi-hunk unified diffs for complex file modifications

## Code Execution

Agents can run shell commands with configurable isolation:

- **Gateway execution**, Run commands directly on the machine hosting the Gateway
- **Sandbox execution**, Run commands in isolated Docker containers with resource limits
- **Node execution**, Run commands on connected companion devices (macOS, Linux nodes)

### Execution Features
- Configurable timeouts (default 30 minutes)
- Background job management (list, poll, kill, clear)
- Auto-background for long-running commands
- TTY support for interactive programs
- Elevated mode for privileged operations (gateway host only)

### Approval Workflows

Destructive command protection explains why a command was stopped and what review is required. Use it to distinguish a safety block from a failed tool call.

![Destructive command guard explaining a blocked command](https://neotask-marketing-assets-417007889150.s3.us-east-1.amazonaws.com/landing/agents/2026-08-12-r1/destructive-guard-1280.webp)
Sensitive commands can require user approval before execution:
- **Allowlist**, Only pre-approved commands execute
- **Ask**, Unknown commands prompt for approval
- **Full**, No restrictions

## Browser Automation

Agents control a full Chromium-based browser (Chrome, Brave, or Edge):

### Actions
- **Navigate**, Open URLs and navigate between pages
- **Interact**, Click, type, hover, drag, select, fill forms
- **Capture**, Take screenshots, export PDFs
- **Inspect**, Snapshot the DOM or accessibility tree
- **Execute**, Run JavaScript in the page context
- **Upload**, Upload files to web forms
- **Manage**, Handle dialogs (alerts, confirmations), read console output

### Browser Profiles
Multiple isolated browser profiles for separating accounts:
- Each profile has its own cookies, storage, and history
- Auto-assigned CDP ports for programmatic control
- Remote browser support (connect to browsers on other machines)

## Canvas & A2UI

The Agent-to-UI (A2UI) system lets agents render interactive visual content on connected devices:

- **Present**, Display web pages or custom content on a device's canvas
- **Navigate**, Control what's shown in the canvas window
- **Evaluate**, Execute JavaScript in the canvas context
- **Snapshot**, Capture what's currently displayed
- **A2UI Push**, Send structured UI updates (JSONL payloads) to the canvas
- **A2UI Reset**, Clear the canvas state

Canvas rendering is available on macOS (WebKit), iOS (SwiftUI), Android (Jetpack Compose), and web (WebChat).

## Node Device Capabilities

Through connected companion apps, agents can:

### Camera
- **Snap**, Take photos (front or back camera)
- **Clip**, Record video clips with configurable duration and FPS

### Screen
- **Record**, Capture screen recordings with configurable duration and quality

### Location
- **Get**, Retrieve GPS coordinates with accuracy selection

### Notifications
- **Notify**, Send native OS notifications with title, body, and priority

### System Commands
- **Run**, Execute shell commands on the node device
- **Which**, Check if a command is available on the node

## Messaging

Agents can send and manage messages across connected channels:

- **Send**, Send messages to any connected channel/recipient
- **Edit**, Edit previously sent messages
- **React**, Add reactions to messages
- **Thread**, Create threads and reply within them
- **Poll**, Create polls on supported platforms
- **Voice**, Check voice channel status (Discord)

## Web

- **Search**, Web search via Brave Search API
- **Fetch**, Retrieve web pages and convert HTML to markdown/text

## Memory & Context

- **Memory Search**, Vector similarity search over agent memory files
- **Memory Get**, Retrieve specific memory files by name

## Session Management

- **List Sessions**, View active sessions across agents
- **Session History**, Fetch transcript for any session
- **Send to Session**, Send messages to other agent sessions
- **Spawn Sub-Agent**, Start isolated sub-agent runs
- **List Agents**, View available agents

## Automation

- **Cron**, Create, edit, run, and manage scheduled jobs
- **Gateway**, Restart gateway, apply configuration changes, run updates

## Image Analysis

Agents can analyze images using configured vision models, describe content, extract text, identify objects, and answer questions about visual content.

## Tool Groups

Tools are organized into groups for easy permission management:

| Group | Tools |
|-------|-------|
| **runtime** | exec, bash, process |
| **fs** | read, write, edit, apply_patch |
| **sessions** | sessions_list, sessions_history, sessions_send, sessions_spawn, session_status |
| **memory** | memory_search, memory_get |
| **web** | web_search, web_fetch |
| **ui** | browser, canvas |
| **automation** | cron, gateway |
| **messaging** | message |
| **nodes** | nodes |