Midokoro for Premiere

Connections: transcription and AI

Midokoro uses two services that you choose: one to transcribe the audio and another, optional, for an AI to review the moments.

Open Settings (the gear at the top right) and scroll down to Connections. There are two cards: Transcription and Moment review (AI). Keys are saved in your computer's secure store, never in a text file.

Transcription

Turns the audio into text with timestamps. Choose the service you already pay for, type or choose the model, and paste your key; then click Test connection.

OpenAI (recommended)	Returns the time of every word and every line and works well in Spanish, English, and many more languages.
Groq	Very fast Whisper. Returns word and line times.
Deepgram	Returns word times.
AssemblyAI	Returns word times.
ElevenLabs	Scribe returns word times.
OpenAI-compatible server	Any server that speaks the OpenAI transcription protocol (a local Whisper, a proxy, or another provider). You choose the address, the path, the model, and the key; it must return the times.

Transcription needs timestamps. A model that returns only text can't be used. The connection test tells you whether the service returns word times, line times only, or none.

Moment review (AI)

An AI reads the transcript and chooses and polishes the moments. We recommend OpenAI, but you can use any AI. The card has three tabs, just like in Synthetic Agents 2.0.

Subscriptions

Use the Claude or ChatGPT plan you already pay for, through its official program: no key and no per-use charge. Each subscription is a card with a switch that starts off.

Claude (Claude Code): uses your Claude plan (Pro or Max). It runs the official Claude Code program, which comes with the Claude desktop app.
ChatGPT (Codex): uses your ChatGPT plan. It runs the official Codex program; keep it up to date.

Turn on the subscription

Click the card's switch. Below each one you see what its terms say today and a link to read them: the decision to use it is yours.

Click Sign in to Claude Code (or to Codex). A window of the official program opens and you finish in the browser. Midokoro never sees your username or password.

Choose the model (optional)

In the drop-down you can leave "Default model of the program" or pick one. With Codex, the list shows the models your account offers. Other… lets you type any model.

Test it

Click Test connection. A subscription can take a few seconds: the program starts a full session.

A subscription is slower than an API: on a long video the search can take several times longer, and the result may vary. If you're in a hurry, use an API; if you'd rather not pay per use, the subscription works just as well.

If the program isn't installed on your computer, the card lets you know.

API keys

Choose the service, type or choose the model, paste your key, and click Test connection. You can use OpenAI (recommended), Groq, OpenRouter, Google Gemini, Anthropic (API), or any other OpenAI-compatible server, with its own address, path, model, and key (under Advanced you can change the authorization header).

Local

With Ollama you use models that run on your computer, free and without a key: install Ollama, download a model, and type its name.

Costs

Midokoro doesn't charge you for transcribing or for the AI: the use of each service is billed to your account with that provider, and there is no per-use charge if you use a subscription or a local model.

If something fails

The connection test messages tell you what happened (key rejected, model that doesn't exist, not signed in…). See Troubleshooting.

Subtitles What ends up in your project