Atlas Commands
Reference guide for all Atlas CLI commands and flags.
Atlas CLI Commands#
All commands are invoked as atlas <command> [flags].
Run atlas --help for a full list at any time.
atlas#
The core command to analyze, build, fix, run tests, and deploy a project.
Flags#
| Flag | Description |
|---|---|
--path <path> | Project path (alternative to positional arg, use this for paths that collide with a subcommand name) |
--model <name> | LLM provider to use (e.g., anthropic, openai) (optional) |
--action <mode> | Action mode: build, test, deploy, test-and-deploy (optional) |
--provider <name> | Deployment provider: vercel, render, netlify, fly, railway (optional) |
--output-dir <path> | Manually specify the static output directory for deployment (optional) |
--allow-dirty | Skip the uncommitted-changes check and proceed anyway |
--auto-rollback-on-unhealthy | Automatically rollback without prompting if post-deploy health check fails |
Interactive Wizard
Running atlas <path> without all required flags will launch an interactive terminal wizard to guide you through selecting a model, picking an action, and setting credentials.
Path Collision Caveat
If your project directory shares a name with a registered subcommand (like models, providers, debug), atlas <that-name> will run the subcommand, not treat it as a path. Use the --path <that-name> escape hatch to disambiguate.
Examples#
Run a complete deployment pipeline silently by providing all flags:
atlas providers#
Manage and view authentication status for all supported deploy providers.
Example Output:
Subcommands#
Securely store a deployment provider token in your OS keychain.
Security Note: Secret values are never printed to the console — only detected, authenticated, or not configured status. Vercel and Render will check their respective env vars (VERCEL_TOKEN, RENDER_TOKEN) before falling back to the credential store.
atlas models#
Manage and view API key status for all supported LLM providers.
Example Output:
Subcommands#
Securely store an LLM API key.
The active: label comes from .atlas/config.json's llm_provider field in the current directory. Only environment variable presence is checked here; use atlas testllm to verify the key actually works.
atlas testllm#
Send a small ping to the configured LLM to verify the API key is active and successfully authenticated.
atlas status#
Show the status and pipeline progress of the latest or a specified Atlas deployment session. It also performs a live health check on the deployment URL if available.
Flags#
| Flag | Description |
|---|---|
--session <uuid> | Specific session UUID to check (default: latest) |
--json | Output status as JSON |
Example Output:
atlas debug run-command#
Run an arbitrary shell command in a workspace and show output. Useful for debugging build detection.
Setting Up Credentials#
LLM Providers (Environment Variables)#
Atlas automatically loads .env from the project's root via godotenv. Set the API key for your chosen provider before running:
Vercel Deployment Configurations#
Option A — Environment Variable (CI/CD) Set your Vercel token securely in your environment variables.
Option B — CLI Interactive Setup Securely store your token in your local OS keychain using Atlas.
Option C — CLI Login Callback If no token is found, Atlas will prompt to launch the Vercel interactive login.
Config File (.atlas/config.json)#
You can define project-level defaults in a .atlas/config.json file.
Supported llm_provider values: anthropic, openai, gemini, mistral, groq, grok, local.
For local models via Ollama: