Get started

Running in a minute.

Install the CLI, connect a model provider, and hand Crayon its first task. Bring your own key — or run fully offline with Ollama.

1

Install

Install globally with npm (or pnpm):

terminal
npm install -g crayon-cli
# or:  pnpm add -g crayon-cli

Prefer not to install? Run it straight away:

terminal
npx crayon-cli@latest chat
2

Connect a model provider

The fastest path is the setup wizard — it asks for your provider, model, and API key, then saves them to ~/.crayon/config.json:

terminal
crayon config

Or set an environment variable for your provider and skip the wizard:

ProviderEnvironment variableNotes
AnthropicANTHROPIC_API_KEYClaude models (recommended)
OpenAIOPENAI_API_KEYGPT models
GoogleGEMINI_API_KEYGemini models
OpenRouterOPENROUTER_API_KEYAny model via OpenRouter
Ollama— none —Local & offline, no key required
~/.zshrc or ~/.bashrc
export ANTHROPIC_API_KEY="sk-ant-..."
# optional overrides:
export CRAYON_PROVIDER="anthropic"
export CRAYON_MODEL="claude-sonnet-4-6"
3

Start coding

Open an interactive session in your project:

terminal
cd your-project
crayon chat

Or fire a one-shot task and let it run:

terminal
crayon run "add a health-check endpoint and a test for it"

Control how much it can do on its own with --mode (or the /mode command in chat):

  • ask — confirm every edit and command (default)
  • auto-edit — apply edits, confirm commands
  • plan — plan only, change nothing
  • auto — run freely, block dangerous commands
  • bypass — no prompts (use with care)
4

Run offline (optional)

Point Crayon at a local Ollama server for zero-cost, fully private runs — no API key:

terminal
ollama run qwen2.5-coder:7b     # pull a model once
export CRAYON_PROVIDER="ollama"
crayon chat
5

Go further

Editor: install the VS Code extension for a chat panel with live reasoning and click-to-source citations.

Tools: connect any Model Context Protocol server to give the agent new capabilities:

terminal
crayon mcp add <name> <command> [args...]
crayon mcp list

Command reference

crayon chatStart an interactive agent session
crayon run "<task>"Run a one-shot autonomous task
crayon configRe-run the setup wizard
crayon initInitialize .crayon/ and index the repo
crayon sessionsList saved chat sessions
crayon chat --resumeResume your most recent session
crayon mcp add <name> <cmd>Add an MCP server
crayon updateUpdate to the latest version