Connect a provider, choose defaults and run your first operation.

Add a provider model

1. Open Settings > Models and choose Add model.
2. Pick a provider from the presets: Anthropic, OpenAI, Google, xAI, DeepSeek or OpenRouter.
3. Enter your API key from that provider when asked. Choosing a preset installs what Grapple needs to talk to the provider and saves the key in one step.
4. Choose the models you want from that provider. They appear in your library with their operations, such as Chat, Images or Video.

The key is stored in Settings and isn't displayed again. Removing a model leaves its connection and key in place for other models that use them.

Use Codex or Claude Code for Ape

If you already use OpenAI's Codex command-line agent or Anthropic's Claude Code, Ape can run through it on your own sign-in instead of an API key.

1. In Settings, install the agent if it isn't already on your computer. Grapple installs the command-line agent, which is separate from any desktop app of the same name.
2. Sign in through the agent's own sign-in. Grapple doesn't see your password.
3. In Settings > Ape, choose one of that agent's models as Ape's conversation model.

Choose which model each job uses

Settings > Defaults sets the model Create offers first for each kind of operation: images, image edits, video, music, sound effects, voice and roto. A job with no default stays visibly unset, and Grapple never swaps in a different model on its own.

For roto, clearing the default returns to the models included with Grapple. To try an alternative roto model, add a compatible EfficientTAM or ViTMatte bundle in Settings > Models and pick it here.

Run an operation in Create

1. Switch to Create (Alt+1) and open the page for the job, such as Images or Video.
2. Choose a model. The form shows that model's inputs; required fields are marked, and options show their allowed values.
3. Describe what you want. To add a reference, choose Add reference and pick a project asset. For video sources, choose the frame or range to use, and crop it if only part matters.
4. Choose Generate. The run appears in the page's history.
5. When it finishes, open the result details, or choose Use on timeline to place it with a track, time and range.

Stop requests cancellation. A remote provider may need time to confirm it, and the run shows its actual state in the meantime.

Connect a chat model on your own computer

A local LLM plugin lets Ape use a model served on your machine, for example by a llama.cpp server. The plugin is a short manifest with no code.

1. Start an OpenAI-compatible Chat Completions server on localhost, 127.0.0.1 or [::1]. Version 1 plugins connect without authentication.
2. Create a folder for the plugin, either in grapple-plugins beside grapple.exe or in the per-user plugin folder shown in Settings, and save a plugin.json in it.
3. Fill in a stable ID, a name, the server's base URL, the exact model ID the server expects, and maxInputTokens. Leave enough of the server's context window free for the model's reply.
4. Choose Reload plugins in Settings, then pick the model for Ape.

{
  "schemaVersion": 1,
  "kind": "llm",
  "id": "my-local-model",
  "name": "My local model",
  "llm": {
    "api": "openai-chat-completions",
    "baseUrl": "http://127.0.0.1:11434/v1",
    "model": "my-local-model",
    "maxInputTokens": 8192,
    "capabilities": {
      "imageInput": false,
      "videoInput": false,
      "audioInput": false
    }
  }
}

Turn on imageInput, videoInput or audioInput only when both the model and the server accept that input. An optional control entry adds one selector to Ape's composer, such as a reasoning effort, and maps it to a request parameter the server understands. Grapple sends plugin requests only to the loopback address you give.

Have Ape configure a model

For a provider without a preset, ask Ape to set it up, or choose the model setup skill with / in the Ape composer. Tell it the provider and the model you want. Ape works from the provider's documentation, declares the model's inputs and outputs, and confirms that Settings reports the model as ready. It never asks for a key in the chat: add the provider's key in Settings when Ape asks.

Charges and data

Provider operations are billed by the provider under your account. Check the provider's pricing for the models you use, especially for video.

A Create run sends the provider only the prompt, settings and references you gave it. Ape's conversation model receives what Ape reads while it works, including project details, transcripts and frames it renders.

When a model isn't available

No model can make images

Nothing configured offers that operation. Add a model with it in Settings > Models.

This model is not available to run

Settings lists what is missing, such as a key, a connection or a local file. Resolve it there and the model becomes available.

The chosen model is no longer configured

The model was removed or changed. Pick another model for the page or for Ape.

A run failed

Read the message on the run before trying again. See troubleshooting for timeouts, rejected inputs and uncertain submissions.