kayetanaYour AI API. On your hardware.
Software and IT service firms build AI into products, support and internal tools. kayetana gives you a REST API with your own keys, limits and usage tracking, with no per-token billing.
What you can build with kayetana.
Docs search
Embed product documentation, sort results with rerank and return answers with citations.
Ticket routing
Assign support tickets to categories with classify, with a confidence score per label.
In-product assistants
Build chat with streaming, JSON mode and function calling into your own UI.
Speech features
Turn recordings into text with transcribe, including timestamps.
Custom models
Fine-tune chat models with LoRA and build classifiers from datasets.
Costs under control
Monthly limit per key, usage per day, key and model, plus response time and error rate.
One API for every task.
Under /v2 you'll find chat, embed, rerank, classify, tokenize, detokenize, transcribe, parse, models, datasets, embed-jobs and finetuning. Authentication with bearer keys, streaming via server-sent events.
- Sample code in Python, TypeScript and curl
- Monthly limit per key, 429 when exceeded
- Embeddings as float, int8, uint8, binary, ubinary
Your customers' data stays with you.
When you run AI for customers, you're responsible for their data. With kayetana, data stays on your server and you stay in control.
- No prompt goes to a cloud service
- Fixed costs instead of per-token billing
- Open models, with a default model per task
- Prompts and responses are never stored
- Can run without internet access
- Updates by script
From Playground to code.
Test in the Playground
Try out your prompt, model and parameters.
Copy the code
Grab ready-made sample code in Python, TypeScript or curl.
Create a key
Generate an API key with a monthly limit for your app.
See the API in action.
In the demo, we take a request from the Playground all the way into your code.
Frequently asked questions
Can I use kayetana for multiple customers?
You can create a separate API key with its own monthly limit for each app or customer and track usage per key.
How does rerank work?
Rerank sorts documents by relevance to a query using embedding similarity with bge-m3. It is not a cross-encoder.
Which models can I use?
kayetana ships with qwen3:4b-instruct, qwen3:14b, gpt-oss:20b, bge-m3 and Whisper large-v3-turbo. You can pull more open models by name from the Ollama library.
Is streaming supported?
Yes. Chat streams via server-sent events, including answers with sources.