Continue CLI (cn) is currently in Alpha
cn is an open-source, modular coding agent for the command line. It provides a battle-tested agent loop so you can simply plug in your model, rules, and tools. cn

Quick start

Make sure you have Node.js 18 or higher installed.
# Install
npm i -g @continuedev/cli

# Interactive mode
cn

# Headless mode
cn -p "Generate a conventional commit name for the current git changes"

Basic usage

Out of the box, cn comes with tools that let it understand your codebase, edit files, run terminal commands, and more (if you approve). You can ask cn to:
  • Fix failing tests
  • Find something in the codebase
  • Execute a refactor
  • Write a new feature
  • And a lot more
Use ’@’ to give it file context, or ’/’ to run slash commands. If you want to resume a previous conversation, run cn --resume.

Headless mode (-p flag)

In headless mode, cn will only output its final response, making it perfect for Unix Philosophy-style scripting and automation. For example, you could pipe your git diff into cn to generate a commit message, and write this to a file:
echo "$(git diff) Generate a conventional commit name for the current git changes" | cn -p > commit-message.txt

Configuration

cn uses config.yaml, the exact same configuration file as Continue. This means that you can log in to Continue Hub or use your existing local configuration. To switch between configurations, you can use the /config slash command in cn, or you can start it with the --config flag (e.g. cn --config continuedev/default-agent or cn --config ~/.continue/config.yaml).

Models

Learn how to add custom models here. Then, you can use the /model slash command to switch between them in cn.

Rules

cn supports rules in the same way as the Continue IDE extensions. You can also use the --rule flag to manually include a rule from the hub. For example, cn --rule nate/spanish will tell cn to use this rule to always speak in Spanish.

Tools

cn supports MCP tools, which can be configured in the same way as with the Continue IDE extensions.

Tool Permissions

cn includes a tool permission system to make sure you approve of the agent’s actions. It will begin with minimal permissions but as you approve tool calls, it will add policies to ~/.continue/permissions.yaml to remember your preferences. If you want to explicitly allow or deny tools for a single session, you can use the command line flags --allow, --ask, and --exclude. For example:
# Always allow the Write tool
cn --allow Write()

# Always ask before running curl
cn --ask Bash(curl*)

# Never use the Fetch tool
cn --exclude Fetch