Create and add an OpenAI API key

Watch this step in the getting-started video, then follow the written steps below:

Play from 0:30 · OpenAI: open the dashboard4 min · English captions

Prerequisites

• An OpenAI Platform account at platform.openai.com. This is separate from a ChatGPT account even when you sign in with the same email.
• A payment method or prepaid credits on the Platform organisation.

A ChatGPT Plus, Pro, or Team subscription does not fund API usage. API billing is entirely separate — and so is your RoughCut purchase, which includes no API credits. This is the single most common reason a valid-looking key fails.

Steps

1. Open the OpenAI API keys page and sign in. RoughCut's welcome tour has an Open OpenAI Platform button that goes to the same place.
2. Create a new secret key. The dialog asks for a name, an optional project and an expiration — pick whatever suits you; a key that expires simply has to be replaced later (see below). Copy it immediately. OpenAI shows the full value only once; if you lose it, create another.
3. Set up API billing. On the Platform, billing belongs to the organisation (open the organisation's settings, then its billing page) rather than to the project you may have chosen for the key. Add a payment method or prepaid credits; OpenAI's prepaid billing guide covers the current flow.
4. In RoughCut, open Settings. On the Basic tab (or Full → LLM) find OpenAI API key, paste the value, and click its Save button. RoughCut confirms it was saved to the Keychain. This button saves only the key.
5. For AI editing on OpenAI: LLM provider is already openai on a new installation. If you had chosen Gemini or Claude before, set it to openai and click Save Configuration at the top of the window — the provider choice is not saved by the key's Save button.
6. Optionally pick a model: next to OpenAI model, Choose Model & Estimate… lists the models your key can reach and estimates the cost of a job. The default is gpt-5.6-sol at medium reasoning; GPT-6 Astra is an optional upgrade. The same window has a Reasoning level picker for the selected model; Default (medium) keeps RoughCut's tested setting, and higher levels are slower and cost more. Use Model fills the model and level; Save Configuration applies them.
7. For Cloud transcription (Pro): leave the provider wherever you like and instead choose ✦ Cloud (OpenAI) in the Transcription control on a job. The two settings are independent.
8. Test with a short clip before committing a long recording.

Expected result

The OpenAI API key field shows a green confirmation with the value masked. Depending on what you set up, either the editing pass runs on OpenAI, or Cloud transcription becomes available on jobs.

When a key expires or is revoked

An expired or revoked key is not an error until RoughCut uses it. The next job that reaches the AI stage then fails with an authentication error naming OpenAI. Create a new key on the Platform, open the OpenAI API key field, click Replace…, paste the new value and click Save. Nothing else needs to change: the provider, model and your jobs are untouched. If you want to confirm a key before a long job, Choose Model & Estimate… → Refresh Models calls OpenAI with the saved key and lists what it can reach.

Cloud transcription needs whisper-1

Cloud transcription sends the job's audio to OpenAI's transcription endpoint and requires word-level timestamps, because RoughCut's cut plan and SRT export are both built from word times.

As of the current release, only whisper-1 returns them. Newer transcription models do not, and RoughCut fails loudly rather than producing a broken cut:

OpenAI model '<model>' returned a transcript without word timestamps, which the cut
pipeline and SRT export require. Per OpenAI's docs only 'whisper-1' supports word
timestamps — set Settings → Full → Transcription → OpenAI transcription model to 'whisper-1'.

Leave OpenAI transcription model at whisper-1 unless you have confirmed a newer model returns word timestamps.

What is sent, and when

Your video is never uploaded in any of these paths. Cloud transcription sends audio only, split at silence boundaries so chunks never break mid-sentence. Full detail in Privacy and data flow.

Common mistakes

• Funding ChatGPT instead of the API. They are different balances.
• Looking for billing inside the project. Billing lives under the organisation's settings; a project only scopes the key.
• Saving the key and expecting the provider to switch. The key's Save stores the key. LLM provider and the model are saved by Save Configuration.
• Losing the key. The full value is shown once. Create a new one rather than hunting for it.
• Changing the transcription model. Anything other than whisper-1 currently fails the job at the timestamp check.
• Assuming Cloud transcription works without Pro. It is a Pro feature and needs your own key — both are required.

Troubleshooting