CLI Overview

The Dexto CLI provides two ways to interact with AI agents:

Mode	What It Is	How to Use
CLI Tool	Terminal commands for managing agents, sessions, and configuration	dexto agents install, dexto setup, dexto agents list
Interactive Mode	Chat session with slash commands for real-time control	/model switch, /mcp add, /search

This guide covers the CLI Tool commands. For slash commands available during chat, see Interactive Commands.

What You Can Do

• Talk to any LLM in your terminal or browser
• Create long-lived AI agents with tools, knowledge, and memories
• Turn any folder into a Dexto workspace with authored agents and skills
• Deploy agents locally or on the cloud
• Build custom integrations with Discord, Telegram, Slack, etc.
• Scaffold new AI applications with dexto create-app

---

Main Command

Basic Usage

# Start interactive session (opens Web UI by default)
dexto

# Start interactive CLI mode
dexto --mode cli

# Run a single prompt (auto-uses CLI mode)
dexto "list files here"
dexto -p "create a snake game"

# Start as API server
dexto --mode server

# Run as MCP server
dexto --mode mcp

Mode Auto-Detection

dexto opens the Web UI by default. When you provide a prompt via -p or as a positional argument, Dexto automatically switches to CLI mode for one-shot execution.

Main Command Options

Flag	Description	Example
-v, --version	Show version	dexto --version
-a, --agent <id|path>	Use agent ID or path to config file	dexto -a nano-banana-agent
-p, --prompt <text>	Run single prompt and exit	dexto -p "list files"
-m, --model <model>	Specify LLM model	dexto -m gpt-5-mini
-c, --continue	Continue most recent conversation	dexto -c
-r, --resume <sessionId>	Resume a specific session by ID	dexto --resume my-session
--mode <mode>	Run mode (web/cli/server/mcp, default: web)	dexto --mode cli
--port <port>	Server port (default: 3000 for web, 3001 for server mode)	dexto --port 8080
--skip-setup	Skip initial setup prompts	dexto --skip-setup
-s, --strict	Require all MCP servers to connect	dexto --strict
--no-verbose	Disable verbose output	dexto --no-verbose
--no-interactive	Disable prompts/setup	dexto --no-interactive
--no-auto-install	Disable auto agent install	dexto --no-auto-install
--auto-approve	Auto-approve all tool executions	dexto --auto-approve

Note: The -a, --agent flag accepts both agent IDs from the registry and paths to agent config files. See the Agent Registry for available agents.

Subcommands

create-app - Scaffold New TypeScript App

Create a new Dexto TypeScript application from scratch.

dexto create-app

This command will:

1. Create project structure
2. Set up TypeScript configuration
3. Prompt for LLM provider and API keys
4. Install dependencies
5. Generate example files

init - Initialize a Dexto Workspace

Turn the current folder into a Dexto workspace with visible authored folders for agents and skills.

# Initialize the current folder
dexto init

# Create a workspace agent (interactive if you omit the id)
dexto init agent
dexto init agent review-agent

# Create and auto-link a subagent to the current primary agent
dexto init agent explore-agent --subagent

# Set or change the workspace primary agent
dexto init primary review-agent

# Create a workspace skill
dexto init skill code-review

# Inspect the current workspace config and deploy preview
dexto init status

What dexto init creates:

• AGENTS.md
• agents/
• skills/
• skills/create-skill/ as a starter skill bundle for authoring more skills

What the follow-up commands do:

• dexto init agent creates agents/<id>/<id>.yml
• The first non-subagent agent becomes the workspace primary by default
• dexto init agent --subagent creates a subagent and links it to the current primary agent when one exists
• dexto init skill <id> creates skills/<id>/SKILL.md plus handlers/, scripts/, mcps/, and references/
• dexto init status shows the current workspace structure, primary agent, skills, and what dexto deploy would use

init-app - Initialize Existing TypeScript App

Add Dexto to an existing TypeScript project.

dexto init-app

Requirements:

• Must have package.json and tsconfig.json in current directory

setup - Configure Global Preferences

Configure global Dexto preferences including default LLM provider, model, and agent.

dexto setup
dexto setup --provider openai --model gpt-5-mini
dexto setup --force

Options:

• --provider <provider> - LLM provider (openai, anthropic, google, groq, xai, cohere)
• --model <model> - Model name (uses provider default if not specified)
• --default-agent <agent> - Default agent name (default: coding-agent)
• --force - Overwrite existing setup without confirmation
• --no-interactive - Skip interactive prompts

See Global Preferences for detailed configuration guide.

deploy - Deploy the Current Workspace to Cloud

Deploy the current folder to a cloud sandbox and keep it linked to that deployment.

# First deploy from any folder
dexto deploy

# List cloud deployments on your account
dexto deploy list

# Open the linked deployment in the dashboard
dexto deploy open

# Inspect the linked deployment
dexto deploy status

# Stop the linked cloud sandbox
dexto deploy stop

# Delete the linked deployment and unlink this workspace
dexto deploy delete

How first-run deploy works:

• Creates .dexto/deploy.json in the current workspace
• Uses the primary workspace agent from agents/registry.json when the workspace defines one
• Falls back to the default cloud agent when the workspace does not define a primary workspace agent
• Uploads the workspace snapshot and links the current folder to the created cloud deployment

Authentication:

• Authenticate with dexto login or set DEXTO_API_KEY

Generated deploy config:

Default cloud agent:

{
  "version": 1,
  "agent": {
    "type": "cloud-default"
  },
  "exclude": [".git", "node_modules", "dist", ".next", ".turbo", ".env*"]
}

Workspace agent:

{
  "version": 1,
  "agent": {
    "type": "workspace",
    "path": "agents/review-agent/review-agent.yml"
  },
  "exclude": [".git", "node_modules", "dist", ".next", ".turbo", ".env*"]
}

Notes:

• dexto deploy deploys the whole workspace, not just git-tracked files
• dexto deploy status, open, stop, and delete work from the linked workspace folder
• dexto deploy list shows account deployments and highlights the one linked to the current folder
• Dashboard links are printed after deploy and status so you can inspect the cloud workspace directly

agents install - Install Agents

Install agents from the registry or custom YAML files/directories.

# Install single agent from registry
dexto agents install nano-banana-agent

# Install multiple agents
dexto agents install podcast-agent coding-agent database-agent

# Install all available agents
dexto agents install --all

# Install custom agent from file
dexto agents install ./my-agent.yml

# Install from directory (interactive)
dexto agents install ./my-agent-dir/

Options:

• --all - Install all available agents from registry
• --force - Force reinstall even if agent is already installed
• --no-inject-preferences - Skip injecting global preferences into installed agents

See the Agent Registry for available agents.

agents uninstall - Uninstall Agents

Remove agents from your local installation.

# Uninstall single agent
dexto agents uninstall nano-banana-agent

# Uninstall multiple agents
dexto agents uninstall agent1 agent2

# Uninstall all agents
dexto agents uninstall --all

Options:

• --all - Uninstall all installed agents
• --force - Force uninstall even if agent is protected (e.g., coding-agent)

upgrade - Upgrade Dexto CLI

Upgrade the Dexto CLI itself. If an older npm global install is detected, this command migrates to native installer automatically.

# Upgrade to latest
dexto upgrade

# Upgrade to a specific version
dexto upgrade 1.6.8

Options:

• --dry-run - Print install/migration commands without executing
• --force - Force reinstall for native install target

uninstall - Uninstall Dexto CLI

Uninstall the Dexto CLI binary (agent removal is handled by dexto agents uninstall).

# Remove CLI binary only (keeps ~/.dexto)
dexto uninstall

# Preview changes
dexto uninstall --dry-run

# Also remove ~/.dexto completely
dexto uninstall --purge

Options:

• --purge - Remove ~/.dexto completely
• --dry-run - Show what would be removed

agents sync - Sync Agent Configs

Sync installed agents with bundled versions after Dexto updates.

# Check status and prompt for updates
dexto agents sync

# List what would change (dry run)
dexto agents sync --list

# Force update all without prompts
dexto agents sync --force

Options:

• --list - Show status without making changes
• --force - Update all agents without confirmation

When to use: When Dexto shows "Agent updates available" notification after an update, or when you want to reset agents to their default configurations.

agents list - List Available Agents

List agents from the registry and locally installed agents.

# List all agents (registry + installed)
dexto agents list

# Show only installed agents
dexto agents list --installed

# Show only registry agents
dexto agents list --available

# Show detailed information
dexto agents list --verbose

Options:

• --verbose - Show detailed agent information
• --installed - Show only installed agents
• --available - Show only available agents from registry

See the Agent Registry for detailed agent information.

which - Show Agent Path

Display the path to a specific agent's configuration file.

dexto which nano-banana-agent
dexto which coding-agent

session - Manage Sessions

Manage conversation sessions.

session list

List all available sessions.

dexto session list

session history

Show message history for a session.

# Show history for current session
dexto session history

# Show history for specific session
dexto session history my-session-id

session delete

Delete a specific session.

dexto session delete old-session-id

search - Search Session History

Search across all conversation messages in session history.

# Search all sessions
dexto search "bug fix"

# Search in specific session
dexto search "error" --session my-session

# Filter by role
dexto search "help" --role assistant

# Limit results
dexto search "code" --limit 20

Options:

• --session <sessionId> - Search in specific session only
• --role <role> - Filter by role (user, assistant, system, tool)
• --limit <number> - Limit number of results (default: 10)

mcp - MCP Server Mode

Start Dexto as an MCP server to aggregate and re-expose tools from configured MCP servers.

# Start MCP tool aggregation server
dexto mcp --group-servers

# Start in strict mode
dexto mcp --group-servers --strict

Options:

• --group-servers - Aggregate and re-expose tools from configured MCP servers
• -s, --strict - Require all MCP server connections to succeed
• --name <name> - MCP server name (default: 'dexto-tools')
• --version <version> - MCP server version (default: '1.0.0')

Note: In the future, dexto --mode mcp will be moved to this subcommand to expose the agent as an MCP server by default.

---

Common Usage Patterns

Quick Start

# Interactive session with default settings (opens Web UI)
dexto

# Interactive CLI mode
dexto --mode cli

# Use a specific agent (opens Web UI)
dexto --agent nano-banana-agent

# Start with a specific model (opens Web UI)
dexto -m claude-sonnet-4-5-20250929

One-Shot Prompts

# Run single task and exit (auto-uses CLI mode)
dexto "list all TypeScript files in src/"
dexto -p "create a README for this project"

# With auto-approve for automation
dexto --auto-approve "format all JavaScript files"

# With specific agent
dexto --agent coding-agent "create a landing page for my coffee shop"

# Combine agent, model, and auto-approve
dexto --agent coding-agent -m gpt-5 --auto-approve "build a todo app with React"

Session Continuation

# Continue most recent conversation (opens Web UI)
dexto --continue

# Continue in CLI mode
dexto --continue --mode cli

# Continue with a one-shot prompt, then exit
dexto -c -p "now add error handling"

# Resume specific session (opens Web UI)
# Get session id from the web UI or session list command
dexto --resume my-project-session

# Resume session in CLI mode
dexto --resume my-project-session --mode cli

# Resume and run a one-shot prompt
dexto -r my-project-session "fix the bug we discussed"

Agent Management

# Install agents for specific use cases
dexto agents install podcast-agent music-agent coding-agent

# Install all available agents
dexto agents install --all

# List what's installed
dexto agents list --installed

# Find agent config location
dexto which coding-agent

# Use custom agent file
dexto --agent ./agents/my-custom-agent.yml

Web UI

# Launch on default port (3000)
dexto

# Custom port
dexto --port 8080

# With specific agent
dexto --agent database-agent

API Server

# Start REST + SSE streaming server (default port 3001)
dexto --mode server

# With custom port
dexto --mode server --port 8080

# With specific agent and strict mode
dexto --mode server --agent my-agent --strict

# For production with custom agent
dexto --mode server --agent ./production-agent.yml --port 3001

Content Generation

# Generate podcast content
dexto --agent podcast-agent "create a 5-minute podcast about space exploration"

# Generate images
dexto --agent nano-banana-agent "create a futuristic cityscape"

# Create code with specific instructions
dexto --agent coding-agent "build a REST API with Express and TypeScript"

# Interactive mode for complex tasks
dexto --agent coding-agent
# Then in the UI: "Let's build a full-stack app step by step"

Automation & CI/CD

# Automated code review (no confirmation prompts)
dexto --auto-approve "review all files in src/ and suggest improvements"

# Generate documentation
dexto --auto-approve "create API documentation from the code in src/api/"

# Run tests and analyze results
dexto "run the test suite and explain any failures"

# Git commit message generation
git diff | dexto -p "generate a conventional commit message for these changes"

Cloud Workspace Deploys

# Deploy any folder with the managed cloud agent
dexto deploy

# List your cloud deployments
dexto deploy list

# Deploy a repo that already defines a workspace primary agent
cd my-project
dexto deploy

# Open the linked deployment in the dashboard
dexto deploy open

# Check the linked deployment later
dexto deploy status

# Stop or delete the linked deployment
dexto deploy stop
dexto deploy delete

Multi-Agent Workflows

# Start researcher agent as MCP server (Terminal 1)
dexto --mode mcp --port 4000 --agent researcher-agent

# Start coordinator agent that uses researcher (Terminal 2)
dexto --agent coordinator-agent --port 5000

Search & History

# Search all conversations
dexto search "database schema"

# Search in specific session
dexto search "bug fix" --session my-session-id

# Filter by role
dexto search "error" --role assistant

# View session history
dexto session history my-session-id

Next Steps

• Interactive Commands - Slash commands for chat sessions
• Agent Registry - Browse available agents
• Agent Configuration - Create custom agents
• MCP Integration - Connect external tools and services
• Global Preferences - Configure system-wide defaults