Working with Ape

Set up Ape

Open Settings > Ape and choose the conversation model. If the list is empty, add a model first in Settings > Models.
For a provider preset such as Anthropic, OpenAI or Google, add the provider and paste its API key. For Codex or Claude Code, Grapple can install the command-line agent; you then sign in with your own OpenAI or Anthropic account through that agent.
Open the Ape panel from the Window menu or the Ape menu.

The model menu in the Ape panel switches models for the current chat, and offers an Effort setting where the model has one. The model setup guide covers providers in detail.

Write a brief Ape can act on

A good brief answers the questions an assistant editor would ask before starting.

What to make: "A lower third", "a four-word kinetic title", "a vertical version of the cut".
Where: The composition, the clip or track, and the time: "over the interview clip, starting at 0:12".
The exact content: The wording of titles, the names in a list, the colours or font to use.
The movement: How things arrive, hold and leave: "rises into place and settles", "hold four seconds, fade out".
What stays: What must not change: "don't touch the edit", "keep the existing grade".
How to check: The range to preview, or what a good result looks like.

Start with a short, well-defined job. It is easier to judge, and the result teaches you how Ape interprets your wording before you hand it something larger.

Use a skill for a longer piece

Type / in the Ape composer and choose /skill to pick one of Ape's skills. A skill is a working method for a type of film, which Ape follows stage by stage.

Documentary opens: Cold opens, title sequences, act openers and chapter cards: archive stills staged as spaces, footage cut on breath and score, typography that behaves like a record.
Kinetic editorial: Fast, music-driven editorial motion from imagery, typography, texture, depth, masks and camera moves.
Scene-based stories: Layered cinematic scenes, with generated artwork made as separate parts and placed at different depths for the camera to explore.
Model setup: Configuring a new provider model so Create and Ape can use it.

The composer also has /settings, which opens Ape's settings, and /compact, which condenses a long conversation when its context window is filling up. The context window indicator in the panel shows how full it is.

Follow and steer a run

Once Ape starts, the plan appears above the conversation and marks each step as it finishes. Expand it to see every step.

To change direction partway through, type a note and press Enter. It goes to the run in progress.
To end the run, choose Stop, which cancels the remaining work. Changes Ape has already made stay in the project, and Undo run can reverse them.
Undo and Redo are paused while Ape is changing the project, so your edits and Ape's don't collide.

With Grapple Ultra turned on in the model menu, Ape can hand independent parts of a large task to private child agents. The Subagents panel lists them, and each can be stopped on its own.

Review what Ape changed

Watch the preview Ape rendered, then play the section yourself.
Switch the conversation to the Changes view for a list of what Ape committed, or to Tool-call for every step in detail.
Choose Review in Stage to compare the frame before and after the run.
In History, the run is a single entry. Expand it to see each change, and choose Show changed to highlight the affected rows in the timeline.

Then either give a specific note, such as "the title sits too close to her face; move it left and start it half a second earlier", or make the change yourself with the editor's tools. Both work on the same layers and keys.

Undo a run

If a run went in the wrong direction, choose Undo run on it in the Ape panel. It reverses everything the run changed in one step.

Undo run isn't available while Ape is still working, or for a run that made no project changes. If you have edited the project since, Grapple checks that your later edits can be kept, and tells you when they can't instead of discarding them. For an earlier state, use Restore before this in History.

Decide what Ape may run

Ape can run terminal commands, for example to inspect files or measure rendered frames. Choose how those commands are approved with the terminal access setting in the Ape panel:

Ask for approval: Every terminal command waits for you. A good default while you learn how Ape works.
Approve for me: Sandboxed commands run automatically, and anything outside the sandbox asks first.
Full access: Commands run with unrestricted filesystem and network access, without asking. Use it only for projects and machines where that is acceptable.

Chats

Choose New chat to start a fresh conversation and Chat history to return to an earlier one. You can't switch chats while a run is in progress; stop it first. Keeping one chat per task makes the history easier to follow.