CLI Guide
This guide helps you get started with the Dexto CLI and includes a comprehensive list of commands you can run with Dexto CLI.
Dexto CLI is the easiest way to get started with AI agents.
Some of the cool things you can do with Dexto CLI:
- Talk to any LLM in your terminal
- Create long-lived AI agents with tools, knowledge and memories. Example: a productivity agent that integrates with your linear and github.
- Deploy these agents either locally or on the cloud
- Talk to these agents on any application - discord, telegram, slack, cursor, claude desktop, etc.
- Start building your own AI applications - get started with building your own Cursor!
dexto create-app
Complete CLI Reference
Main Commands & Options
| Command | Description | Example |
|---|---|---|
dexto | Start interactive CLI | dexto |
dexto "prompt" | Run single prompt | dexto "list files here" |
dexto -p "prompt" | Run single prompt (explicit) | dexto -p "create a file" |
dexto --mode web | Launch web UI | dexto --mode web --web-port 8080 |
dexto --mode server | Start as API server | dexto --mode server |
dexto --mode mcp | Run as MCP server | dexto --mode mcp |
dexto --mode discord | Start Discord bot | dexto --mode discord |
dexto --mode telegram | Start Telegram bot | dexto --mode telegram |
CLI Flags & Options
| Flag | Description | Example |
|---|---|---|
-v, --version | Show version | dexto --version |
-a, --agent <path> | Use custom agent config | dexto -a ./my-agent.yml |
-m, --model <model> | Specify LLM model | dexto -m claude-4-sonnet-20250514 |
-r, --router <router> | Specify router (vercel/in-built) | dexto -r in-built |
--new-session [id] | Start new session | dexto --new-session my-session |
--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 -p "format my repo" |
--web-port <port> | Web UI port | dexto --mode web --web-port 3001 |
Subcommands
| Command | Description | Example |
|---|---|---|
create-app | Scaffold new Dexto TypeScript app | dexto create-app |
init-app | Add Dexto to existing TypeScript app | dexto init-app |
setup | Configure global preferences | dexto setup |
install [agents...] | Install agents from registry | dexto install nano-banana-agent |
uninstall [agents...] | Uninstall local agents | dexto uninstall nano-banana-agent |
list-agents | List available/installed agents | dexto list-agents --installed |
which <agent> | Show path to agent | dexto which nano-banana-agent |
mcp --group-servers | Start MCP server aggregator | dexto mcp --group-servers |
Interactive CLI Commands
Once in interactive mode (dexto), use these slash commands:
| Command | Description | Example |
|---|---|---|
/help [command] | Show help information | /help session |
/exit, /quit, /q | Exit CLI | /exit |
/clear, /reset | Clear conversation history | /clear |
/session list | List all sessions | /session list |
/session create [id] | Create new session | /session create work-session |
/session load <id> | Load session | /session load work-session |
/session delete <id> | Delete session | /session delete old-session |
/session export <id> | Export session data | /session export work-session |
/history [limit] | Show conversation history | /history 10 |
/search <query> | Search conversation history | /search "code review" |
/model list | List available models | /model list |
/model switch <model> | Switch LLM model | /model switch gpt-4o |
/model current | Show current model | /model current |
/mcp list | List MCP servers | /mcp list |
/mcp connect <name> | Connect MCP server | /mcp connect filesystem |
/mcp disconnect <name> | Disconnect MCP server | /mcp disconnect web |
/mcp status | Show connection status | /mcp status |
/tools list | List available tools | /tools list |
/tools search <query> | Search tools | /tools search file |
/prompt show | Show current system prompt | /prompt show |
/prompt reload | Reload system prompt | /prompt reload |
/log level <level> | Set log level | /log level debug |
/log tail [lines] | Show recent logs | /log tail 50 |
/config validate | Validate configuration | /config validate |
/config reload | Reload configuration | /config reload |
/stats | Show system statistics | /stats |
Common Usage Patterns
See all available options and flags:
dexto --help
Launch the interactive CLI:
dexto
Start dexto CLI with a different LLM
# openai
dexto -m gpt-5-mini
# anthropic
dexto -m claude-sonnet-4-5-20250929
# google
dexto -m gemini-2.5-flash
Start dexto with a different config file
This allows you to configure dexto CLI to use a different AI agent
dexto --agent <path_to_agent_config_file>
Check Configuration Guide to understand more about dexto config files
Require all MCP servers to connect successfully
By default, Dexto uses "lenient" mode where individual servers can fail to connect without preventing startup. Use the --strict flag to require all servers to connect successfully:
dexto --strict
This overrides any individual connectionMode settings in your MCP server configurations. See MCP Configuration for more details on connection modes.
Skip tool confirmation prompts during development
dexto --auto-approve "refactor my project using the filesystem and browser tools"
Use the --auto-approve flag when you trust the tools being triggered and want to bypass interactive confirmation prompts. This flag overrides the toolConfirmation.mode defined in your agent config for the current run only.
Run a specific command with Dexto CLI:
dexto "find all .sh files in this directory"
# or use explicit -p flag
dexto -p "find all .sh files in this directory"
or do the same with gemini:
dexto -m gemini-2.0-flash "find all files in this directory"
# or with explicit -p flag
dexto -m gemini-2.0-flash -p "find all files in this directory"
Dexto CLI can accept any command - if it doesn't see it as an in-built command, it will fire a single run CLI with that request
For instance, in the above command, the query "find all .sh files in this directory" will start Dexto Agent, send it this query, process the response, and then exit.
Start a telegram bot
dexto --mode telegram
To use a specific agent config file for the telegram bot:
dexto --mode telegram --agent ./telegram-agent-config.yml
Start a discord bot
dexto --mode discord
To use a specific agent config file for the discord bot:
dexto --mode discord --agent ./discord-agent-config.yml
Start dexto as an MCP server
dexto --mode mcp
With this, you can now connect this agent to Cursor, claude desktop, or even other Dexto agents!
Check Dexto Agents as MCP Servers to understand more about MCP servers.
Group MCP servers with dexto
dexto mcp --group-servers
This starts Dexto as an MCP server that aggregates and re-exposes tools from multiple configured MCP servers. This is useful when you want to access tools from multiple MCP servers through a single connection.
To use a specific config file:
dexto mcp --group-servers -a ./dexto-tools.yml
Check Using Dexto to group MCP servers to understand more about MCP server aggregation.
Environment variables
Dexto reads settings from the layered environment loader (process.env, .env, .dexto/.env). After you run dexto setup, your LLM credentials live in ~/.dexto/.env, so you rarely need to touch these variables unless you are customizing runtime behaviour.
General runtime controls
| Variable | Default | Purpose |
|---|---|---|
DEXTO_LOG_LEVEL | info | Controls log verbosity (error, warn, info, debug, silly). |
DEXTO_LOG_TO_CONSOLE | false | Force console logging even when logs are redirected to file (set to true). |
DEXTO_ANALYTICS_DISABLED | false | Opt out of analytics when truthy (1, true, yes). |
Analytics overrides
| Variable | Default | Purpose |
|---|---|---|
DEXTO_POSTHOG_KEY | Built-in public key | Supply a custom PostHog project key. |
DEXTO_POSTHOG_HOST | https://app.posthog.com | Point analytics to a self-hosted PostHog instance. |
Web UI & server configuration
| Variable | Default | Purpose |
|---|---|---|
HOSTNAME | 0.0.0.0 | Hostname used when launching the Web UI. |
FRONTEND_PORT | 3000 | Overrides the Next.js port when running --mode web. |
PORT | 3000 | Fallback port used by the API server if API_PORT is not set. |
API_PORT | 3001 | Controls the REST/WebSocket API port for web/server modes. |
API_URL | http://localhost:<API_PORT> | Overrides the API base URL passed to the Web UI. |
FRONTEND_URL | http://localhost:<FRONTEND_PORT> | Overrides the URL opened in the browser when the Web UI starts. |
NEXT_PUBLIC_API_URL | Derived from API_URL | Injects a custom API URL into the Web UI bundle. |
NEXT_PUBLIC_WS_URL | ws://localhost:<API_PORT> | Injects a custom WebSocket URL into the Web UI bundle. |
NEXT_PUBLIC_FRONTEND_URL | Derived from FRONTEND_URL | Injects a custom frontend URL into the Web UI bundle. |
DEXTO_BASE_URL | http://localhost:<PORT> | Base URL used by the REST server when generating absolute links. |
DEXTO_MCP_TRANSPORT_TYPE | http | Switch between http and stdio transports for the MCP aggregation server. |
Discord integration
| Variable | Default | Purpose |
|---|---|---|
DISCORD_BOT_TOKEN | — | Required to enable --mode discord. |
DISCORD_RATE_LIMIT_ENABLED | true | Disable Discord rate limiting by setting to false. |
DISCORD_RATE_LIMIT_SECONDS | 5 | Cooldown window for rate limiting in seconds. |
Telegram integration
| Variable | Default | Purpose |
|---|---|---|
TELEGRAM_BOT_TOKEN | — | Required to enable --mode telegram. |
TELEGRAM_INLINE_QUERY_CONCURRENCY | 4 | Maximum concurrent inline queries handled by the bot. |
Change log level on the fly
DEXTO_LOG_LEVEL=debug dexto what is the time
Project setup commands
These commands will help you get started creating your own AI application using Dexto
Setup a fresh typescript project using dexto-core
dexto create-app
Add dexto into an existing typescript project
dexto init-app
Check Building with Dexto Guide for more information!
Agent Management
Install Pre-built Agent Templates
Dexto provides ready-to-use agent templates from the registry:
# List available agents
dexto list-agents
# Install specific agents
dexto install nano-banana-agent podcast-agent database-agent
# Use an installed agent
dexto --agent nano-banana-agent "create a futuristic cityscape"
dexto --agent podcast-agent "generate a podcast intro"
# Find where an agent is installed
dexto which nano-banana-agent
# Remove agents you no longer need
dexto uninstall nano-banana-agent
Available Agent Templates
- Nano Banana Agent – Advanced image generation and editing using Google's Nano Banana (Gemini 2.5 Flash Image)
- Podcast Agent – Advanced podcast generation using Google Gemini TTS for multi-speaker audio content
- Database Agent – Demo agent for SQL queries and database operations
- Image Editor Agent – Image editing and manipulation
- Music Agent – Music creation and audio processing
- PDF Agent – Document analysis and conversation
- Product Researcher – Product naming and branding research
- Triage Agent – Demo multi-agent customer support routing system
See the full list with dexto list-agents and examples in the agents/ folder.