Models and Providers

Atlas assembles product and project context, then uses your selected model to answer or generate source. Configure your connection in Atlas Providers, then choose the active provider and model from the chat dropdown. Changing the Provider dropdown in the settings window only changes which settings are displayed.

Choose a provider route

Current Atlas services allow every signed-in account to configure its own OpenAI or Claude key, without a separate enablement step. An older service may still return a local-Codex-only catalog. A managed Opsive subscription is not required for local Codex, Claude Code, or your own API key; managed-plan availability is separate.

Set up local Codex

1. Install and sign in to the Codex app or standalone CLI on the same computer as Unity. Follow OpenAI’s authentication instructions.
2. Open Settings > AI providers and API keys… and set Provider to Codex.
3. Set Run with to Auto (CLI, then desktop app), Codex desktop app, or Codex CLI.
4. For a standalone CLI that is not found automatically, set CLI executable to its executable path.
5. Choose Check Runtime & Sign-in and resolve any reported installation or login issue.
6. Return to the chat and choose an available local Codex model from the model dropdown.

The automatic route tries the standalone CLI before the desktop app. Choose a route explicitly if more than one installation exists and you want a particular one. Atlas does not need you to paste Codex login tokens into Unity.

OpenAI distinguishes ChatGPT sign-in from API-key authentication. Use the intended account and authentication method; a ChatGPT subscription and separately billed API usage are different routes.

Set up Claude Code

Atlas can run your own Claude Code installation on Windows or macOS. You do not need to add an Anthropic API key to Atlas for this route. Claude Code uses its own configured sign-in method, which may have subscription or provider usage limits and charges.

1. Install Claude Code, or use a supported Claude desktop app installation that includes the Claude Code runtime.
2. Open Settings > AI providers and API keys…, then select Claude in Atlas Providers.
3. Set Run with to Auto, Claude desktop app, or Claude CLI. If the standalone executable is not discovered, set CLI executable to its path.
4. Choose Check Runtime & Sign-in. If sign-in is required, choose Sign In to Claude Code and complete Claude Code’s own sign-in in the terminal window it opens.
5. Check the status again, return to the chat, and select an available Claude Code model. Viewing provider settings alone does not switch the active chat model.

Signing in to the Claude desktop app does not automatically sign in the Claude Code runtime it installs. The runtime has a separate sign-in; Sign In to Claude Code launches that flow without asking you to copy credentials into Atlas.

Atlas runs the installed program without copying it and does not read or upload its credentials. These local runtime calls do not consume Atlas model credits. The account used by Claude Code determines available models, limits and billing. Claude and Claude API are separate choices: the latter uses an Anthropic API key stored by Atlas.

Add an API key

Create a key with your provider

Create the key in the provider account you want to use for Atlas:

For Claude, a key scoped to one workspace can be used without entering a workspace ID in Atlas. If the key is not scoped to a workspace, copy that workspace’s ID from the Console and enter it in Atlas’s Workspace ID (optional) field. Anthropic’s workspace selection instructions explain which keys require it.

Your provider account controls API access, available models, billing, and usage limits. A saved key must have access to the model you select; successful key validation does not guarantee that the account has enough quota for generation.

Save the key in Atlas

1. Sign in to Atlas, then open Settings > AI providers and API keys… in the Atlas window.
2. In Atlas Providers, set Provider to Codex API or Claude API, matching the account that issued your key.
3. Paste the complete key into New API key. For Claude, also enter Workspace ID (optional) when your key requires it.
4. Choose Save Key, or Replace Key if that provider already has a saved key.
5. Wait for API key validated and saved. The model list has been refreshed. The provider settings then show Using your key with a masked identifier. Atlas checks the key before replacing an existing credential; if validation fails, the previous key is kept.
6. Return to the chat and select a model from that API provider in the model dropdown. Saving a key or viewing its settings does not switch the active chat model.
7. Send a short question to confirm the selected model can answer with your key. If the provider reports a quota or billing error, check that provider account’s API usage allowance.

This setup uses the Atlas settings window; you do not need to put the key in your Unity project files or set an environment variable for Atlas.

Replace or remove a key

To replace a key, open the same provider settings, enter the new value in New API key, and choose Replace Key. To stop Atlas using the saved key, choose Remove Key and confirm the removal. Use Refresh to retrieve the current saved-key status.

Adding a key does not remove the local Codex or Claude Code options. Atlas sends the key over HTTPS and stores it encrypted on the server; Unity does not retain the entered key. Removing it from Atlas does not revoke it at OpenAI or Anthropic. Manage or revoke the provider key in that provider’s dashboard when needed.

Never attach a key to a chat or include it in project instructions. Provider limits, available models, and billing are controlled by the provider account.

Model and effort

The chat model picker selects both provider and model. Atlas initially prefers local Codex when it finds a local runtime, and retains an explicit provider choice. If that route becomes unavailable, check its settings and choose an available model; Atlas does not silently replace an explicit choice with another provider.

The model picker shows choices returned for the current connection. Effort choices vary by model; higher effort may take longer and consume more provider resources. Start with the default, then increase it for a difficult design or debugging task if needed. A model name visible in an example image is not a promise of availability for every account.

Use /model and /effort for keyboard selection. See Commands and shortcuts, Privacy, and Troubleshooting.

Unity project tools

Local Codex can inspect the live Editor on Unity 6 or newer when Unity CLI and Unity Pipeline are installed. Inspection is enabled by default when those dependencies are available. Scene editing has a separate per-project opt-in and requires an explicit request to change the project. See Providing context for setup, supported edits, and limitations.

Connect an assistant to Atlas

You can also use Atlas from an external AI assistant through the Atlas plugin. Its tools and skills provide project inspection, product knowledge, validation and preview/apply workflows. This is a separate connection from choosing the model inside the Atlas window. See AI assistants and skills.