Custom API

A custom API lets kk use provider keys that you obtain yourself. kk is free, you pay providers directly, and audio and text travel only between your computer and the providers you choose, never through kk’s servers.

Roles to configure

kk splits its features into roles, and each role can use a different provider and model.

Role Required What it does
Speech recognition Yes Turns the class audio into English text in real time
Translation Yes Translates the text into your chosen language. It can use the recognition provider’s built-in streaming translation, where offered, or a large language model
Stage analysis and titles Optional Produces stage highlights, session titles and category suggestions, and by default also writes post-class notes. These features are unavailable if it is not configured
Class Q&A Optional Answers your questions about the class. You can add several models and switch between them in the input box. It reuses the stage-analysis model by default

Recognition and translation are enough to start translating; add stage highlights, Q&A and notes as needed.

Setup steps

  1. Get a key. Create an API key in the provider’s console. Settings shows a short note on where to find it under each provider.
  2. Enter it in Settings. Open Settings, choose a provider under “Speech recognition”, enter the key and select “Save key”. Then choose providers and models for “Translation” and “Stage analysis and titles”. A provider’s key only needs to be saved once.
  3. Test the connection. Each role has a “Test connection” button: success shows a green dot and the time taken, failure a red dot and the reason. For language models the time is the first-token delay; for recognition services it is the time to establish the connection.

Keys are stored only in Windows Credential Manager. They are not written to the project folder, not uploaded and not synced to the cloud. Changing providers never sends the old provider’s key to the new one.

Available providers

Speech recognition

AssemblyAI, Deepgram, Speechmatics, Gladia, ElevenLabs, OpenAI, Groq, Tencent Cloud, iFLYTEK, Alibaba Cloud Model Studio and Volcengine. Languages, term hints and translation support vary between services; Settings shows the details once you pick one.

Translation, stage analysis and Q&A

DeepSeek, Claude, Google Gemini, OpenAI, OpenRouter, SiliconFlow, Volcano Ark, Kimi, Qwen, Zhipu, and Ollama running on your own computer.

The list in the app is authoritative and grows with new versions. If your provider offers an OpenAI-compatible endpoint, choose “Add provider…” at the bottom of the provider menu and enter a name and endpoint address (starting with https://).

Built-in providers supply only a name and endpoint, and kk does not pre-fill a model for you: pick one from the provider’s model list or type a model name.

Language and terms

  • “Translation language” sets the language of the caption translation as well as stage highlights, titles, notes and Q&A: Chinese, English, Japanese or Korean.
  • “Class language” tells the recognition service which language the teacher mainly speaks. Some services only recognize English, in which case Settings says so.
  • When the recognition service supports term hints, kk sends the specialist words you added or confirmed yourself to reduce mishearing; see Glossary.

Cost and data

  • Providers bill you at their own prices. kk charges nothing and cannot show your provider bill.
  • Audio goes straight from your computer to the speech recognition provider you chose, and text goes to the language-model provider you chose. See each provider’s own privacy policy.
  • With a custom API, kk does not upload your class content to the official service; see Data and telemetry.

Feature differences

A custom API supports live captions, the floating window and caption bar, stage highlights, the library, notes, the glossary and audio/video import. Cloud sync, the web version and slide following are currently available only to kk Pass users.

Last updated: November 15, 2026