Set up an AI assistant — OriginCut Docs

01Find the assistant settings

Open Settings and go to the Agents area. You can also make this choice during onboarding. Start with one working assistant and switch later when needed.

Detected as installed does not mean signed in, or that the account has an available model. Send a simple request after connecting to confirm it works.

Choose an assistant in Settings, sign in, and check its connection status.

02Connect Codex

Select a working local Codex installation and check its status. If it cannot start or is incompatible, use Download Codex component in Settings.
For the managed component, choose an authorization mode: use the local signed-in subscription, sign in to ChatGPT separately, or configure a custom API.
Separate sign-in opens your browser. A custom API needs the base URL, model ID, and API key required by your provider. Save, then send a message to verify access.

03Prepare another assistant

Claude Code: install it, sign in, and confirm it works independently, then return to OriginCut and detect again.
OpenCode: install OpenCode CLI, configure an available model, then detect and select it.
WorkBuddy: install a supported desktop version. If OriginCut asks you to sign in, complete browser sign-in from the WorkBuddy settings inside OriginCut.

04Confirm the connection

After selecting a project, ask the assistant to list and summarize its media. Once it can reply and read the project, begin editing. Review the specific action when an approval request appears.

Try asking
Inspect the media in this project and suggest two editing directions. Do not change the timeline yet.

05If detection or connection fails

Check the assistant’s installation, sign-in, network access, and account allowance.
Run detection again in Settings. For an incompatible Codex launch, check the managed component option.
Expand provider errors to read their specific message. Keep the diagnostic ID and use in-app feedback if the problem continues.