> ## Documentation Index
> Fetch the complete documentation index at: https://docs.archetype.ing/llms.txt
> Use this file to discover all available pages before exploring further.

# Accounts and keys

> Sign in to each engine, add API keys, choose the model, and bring your setup from other tools.

## Sign in to an engine

Open **Manage → Agents and models**. Each engine has a row that says whether it is installed, signed in, and on which plan.

* **Install** installs an engine that is missing.
* **Sign in** runs the engine's own sign-in. For Claude Code, Codex and Kimi Code it opens a browser; for Antigravity, run `agy` in a terminal. OpenCode starts on its free models without signing in.
* **Use** makes it the engine new conversations start with. A row that is not installed or signed in has none.

Each engine keeps its sign-in in its own files; Archetype does not read them. **Add account** keeps another account of the same engine (a work and a personal one, say), and you choose which one runs.

On a server with no browser, give Claude Code a token from `claude setup-token` as `CLAUDE_CODE_OAUTH_TOKEN` in the env file.

## API keys

Archetype's own models run on your API keys, or on a ChatGPT account: **Sign in** on the OpenAI row shows a code to enter in your browser.

On a provider's row, **Add key** saves a key; **Add another key** adds one more, tried in turn when one is refused or at its limit. Keys go in `~/.config/archetype/env`, one per line:

```bash theme={null}
ANTHROPIC_API_KEY=…
OPENAI_API_KEY=…
GEMINI_API_KEY=…        # or GOOGLE_API_KEY, from Google AI Studio
OPENROUTER_API_KEY=…
```

* Only you may read the file (`chmod 600`); the background service refuses to start otherwise.
* A variable already set in your shell wins over the file.
* The background service reads the file again within 2 seconds of a change.

Gemini works with an API key only; signing in with a Google account is no longer offered.

When the agent needs a key for something else, it asks for it on a card with a hidden field. The key never enters the conversation; it is kept in `~/.local/share/archetype/secrets/`.

## Choose the model

> [tab] App
>
> The chip under the message box chooses the engine and its model, and the reasoning effort. See [Engines](https://docs.archetype.ing/en/use/engines.md).

> [tab] Terminal
>
> ```bash theme={null}
> /backend                     # choose what runs the conversation
> /backend engine:claude       # an engine, or a provider of Archetype's own models
> /model                       # choose a model with a key
> /model NAME
> ```
>
> At the start: `archetype --backend claude`, `--provider anthropic`, `--model NAME`. `archetype --list-models` lists the models.

The conversation goes on after a switch. A default `provider` and `model` go in your user config; a project's config cannot set them.

**Profiles** keep settings to apply together: engine, model, effort, permission mode and Tokenmaxxing. Save the current ones in **Manage → Profiles**, and switch from the status bar or with `/profile NAME`.

## Bring your setup from other tools

The first run offers **Bring your setup**; later it is **Manage → Import environment**. Archetype finds what is on this computer and you choose what to bring:

| From | What comes |
| - | - |
| Claude Code, Codex, Hermes, Cline, OpenCode, Goose, Aider | Instructions, skills and MCP servers |

* Instructions are added to the end of your `AGENTS.md` under each tool's name, as written; nothing is summarized.
* A key an MCP server needs goes to the env file, not the config.
* Not brought: models and providers, API keys, sign-ins, permissions, hooks and history. Sign in again in Archetype.
* Nothing runs while it looks. Imported MCP servers start when you press **Apply now**.

To go on with a Claude Code conversation from this folder, type `/import` in a conversation: it continues in a new session from a copy.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.