How do I make a custom GPT that transcribes audio?
Create a custom GPT, import https://console.kolwrite.com/openapi.json as an Action, set your key as the x-api-key header and paste the instructions below. The GPT can then transcribe, subtitle and translate media files that are online, on your KolWrite credits. An Action sends only JSON, so it works with links to media files, not with files you attach in the chat.
What you do once
Create the account and the key yourself at console.kolwrite.com: sign up, add credits, then open API Keys and create a key. An agent cannot do this for you.
You will paste the key into the GPT editor, not into a chat.
Create the GPT and its Action
- In ChatGPT open Explore GPTs, then Create, then the Configure tab. Building GPTs needs a paid ChatGPT plan.
- Under Actions choose Create new action, then Import from URL, and enter
https://console.kolwrite.com/openapi.json. - Authentication: API Key, Auth Type Custom, Custom Header Name
x-api-key, and paste your KolWrite key. - Privacy policy:
https://www.kolwrite.com/tos_pp.html. - Paste the instructions below into the Instructions field.
- Share: Only me. Anyone who can use the GPT spends your credits.
Instructions to paste
You transcribe audio and video with the KolWrite API actions.
- The person gives a direct link to a media file (mp3, m4a, wav, mp4, mov...). A YouTube or other web page link is not a media file: say so and ask for the file's own link.
- Start with createTranscription. Always send model: "oracle" unless the person asks for cheaper ("base") or hardest audio ("oracle_max"). For subtitles send output_format "srt" and word_timestamps false. For "who said what" send include_diarization true. To translate, send translate_to with the language code.
- After you start a job, tell the person the jobId and call getTranscription. If status is PENDING or IN_PROGRESS, say it is still running and check again when the person asks.
- When status is COMPLETED, show the result (for srt, in a code block) and the creditsUsed value. When FAILED, show the error text as it is.
- Never start the same job twice: each start is billed. If an action call fails, show the error; do not retry createTranscription.
- Tell the person the expected cost before transcribing more than an hour: base US$0.48, oracle US$1.44, oracle_max US$3.60 per hour of audio.
What to ask
Three things to ask the GPT (with your own links):
Transcribe https://example.com/episode-12.mp3
Make English SRT subtitles for https://example.com/talk-he.mp4
Who speaks in https://example.com/call.wav, and what did each person agree to?
What works and what does not
- Works: links to media files that are reachable from the internet, any of the 100 languages, subtitles, speaker labels, translation.
- Does not work: files you upload into the chat. The Action cannot pass them to KolWrite. Use a link, or Claude Code, Codex or kolwrite.py for local files.
- Long recordings take minutes. The GPT cannot wait in the background; ask it to check the job again.
- Very large JSON results may be more than a GPT reply can show; ask for txt or srt instead.
What it costs
| Job | Model | Cost |
|---|---|---|
| A 30-minute interview, plain text | base | US$0.24 |
| A one-hour meeting with speaker labels | oracle | US$1.56 |
| A one-hour Hebrew talk with English subtitles | oracle | US$1.44 |
| One hour of noisy, overlapping audio | oracle_max | US$3.60 |
Billed per second of audio from your credits, with no minimum. Speaker labels add 0.10 credits per audio hour. Full rates: /models.
What the GPT must not do
- Start the same transcription again after a timeout or a network error. Each start is a new job that is billed; it should check the job instead, or ask you.
- Send a YouTube or other web page address as the media. It needs the media file itself.
- Transcribe long recordings (over an hour) without telling you the expected cost first.
- Be shared beyond you while it holds your key.
Questions and answers
Is this the same as the KolWrite connector for ChatGPT?
No. This is a GPT you build with your API key and credits. The connector for a person's own KolWrite account is described in agent-instructions.md.
Can I use the same openapi.json in other tools?
Yes. It is OpenAPI 3.1 with examples and the real error responses; any tool that imports OpenAPI and sends an x-api-key header can use it.