Frequently asked questions

kayetana is a self-hosted AI platform for businesses. It runs on a Mac Studio in your office or on your network and includes chat, answers from documents, search, classification, transcription and fine-tuning. You use it in the browser or through a REST API.

For teams that want to use AI without handing data to cloud AI services. The package includes unlimited users and requests, so your team size doesn't affect the price.

You chat with open models, ask questions about your documents and get answers with a source for each sentence. You can also sort texts by label, transcribe recordings, rank documents by relevance and fine-tune your own models.

Yes. You only need the internet to download models and updates. After that, everything runs entirely on your own network.

Request a demo. In 30 minutes we'll show you the platform with your own examples.

A Mac Studio with Apple silicon. We recommend 36 GB of memory or more so models up to about 20 billion parameters can run alongside search and transcription. More memory fits larger models.

qwen3:4b-instruct, qwen3:14b and gpt-oss:20b for chat and reasoning, bge-m3 for search, reranking and classification, and Whisper large-v3-turbo for speech. All are openly licensed (Apache 2.0 or MIT).

Yes. You load new open models by name from the Ollama library. You set a default model for chat, embed and rerank, and delete models you no longer need.

For answers from documents: PDF, Word (docx/doc), RTF, ODT and HTML, or pasted text. For audio: MP3, M4A, WAV, AIFF, FLAC and more, up to 100 MB. Datasets can be uploaded as CSV, JSONL, TXT, PDF or Word.

Yes. You fine-tune a chat model (currently qwen3:4b-instruct) with LoRA on your own example conversations and get a model named “ft-name”. From labeled datasets, you build classifiers named “clf-name” on top of an embedding model.

You install updates with a script, and all services start on their own after a reboot. Every night at 3:15 a.m., kayetana backs up the database and files and keeps the backups for 14 days. We also recommend an external backup.

Only you and your team. Models, documents and logs live on your own machine, and no prompt goes to a cloud provider.

Prompts and answers are never stored. Only time, endpoint, model, status, tokens and duration are logged. You can delete the log at any time.

You choose a retention period for usage logs and datasets: unlimited, 7, 30, 90 or 365 days. After that, kayetana deletes the data automatically.

Yes. kayetana uses HTTPS on your own network with its own certificate. The application itself only listens locally.

There are Owner, Admin and Member roles, and you invite new people by link. A sign-in lasts 14 days, and after 10 failed attempts the account is locked for 15 minutes.

No, we don't hold any such certification. kayetana supports you with data protection because your data stays in the building. That is not a guarantee or legal advice.

The REST API under /v2 offers chat, embed, rerank, classify, tokenize, detokenize, transcribe, parse, models, datasets, embed-jobs and finetuning. That gives your own apps access to every feature of the platform.

With an API key sent as a bearer token. You set a monthly limit per key, and once it's exceeded, the API returns status 429.

Yes. Chat answers arrive piece by piece via server-sent events. Answers from documents with citations can be streamed too.

You pass your texts in the documents parameter. The response then includes citations that map each statement to its source.

Yes. The Playground shows ready-to-use code in Python, TypeScript and curl for every request. You can copy it straight into your app.

Yes. Embed jobs process entire datasets in the background. You can follow the progress, and the result is available to download as JSONL.

Pricing is available on request. It's a flat fee for the platform, with no per-token or per-user billing.

All features, unlimited users and requests, on-site or remote setup, updates for the platform and models, nightly backups, support by email and phone, plus tutorials and onboarding. The API comes at no extra charge.

You can use your own Mac Studio or have us supply one. We'll advise you on how much memory makes sense for your tasks.

You set up the Mac Studio, and all it needs is power and a network connection. We load the right models and set up HTTPS and backups, and you invite your team. Then you get going in the browser or through the API.

There are short videos and step-by-step guides for every feature. On top of that, you get onboarding during setup.