AI that stays on your machines.
Chat, documents, images and video on hardware you own — every run priced up front and recorded.
Recorded on dev.orazaka.com with a demo account — nothing staged.
- 128
- runs measured on real hardware, each with its record
- 9
- models measured, on a Mac or hosted
- 0
- telemetry calls — nothing phones home
What it does
Three things, done on your side.
Every screen below is the real app.
Studios, priced before they run
A Studio turns a sentence or a file into an image, a video or a checked document. Its estimated cost shows before it starts.
Every result keeps its run
The output, its steps, its input and its run id stay together — so any result can be checked again.
Chat and agents that ask first
Models run on your machines; a hosted one only if you connect it. An agent shows its plan and acts once you approve it.
Measured
One prompt, three models, three run records.
Times, costs and run ids come from the platform's own records — they are not typed.
![A winter night in Montréal, by FLUX.2 [klein] 4B](https://krizaka-assets.b-cdn.net/site/products/orazaka/compare/montreal-winter--flux-2-klein-4b.c141d84ef2.webp)
FLUX.2 [klein] 4B
- Ran on
- One Mac (Metal) — nothing left the machine
- Time
- 66 s, from the command to the file
- Provider price
- None — your own machine
- Run
- 6a4c4310
Stable Diffusion 1.5
- Ran on
- One Mac (Metal) — nothing left the machine
- Time
- 58 s, from the command to the file
- Provider price
- None — your own machine
- Run
- 5b4ef17f
Stable Diffusion 3.5 Large
- Ran on
- DigitalOcean serverless inference
- Time
- 8 s, from the command to the file
- Provider price
- $0.08 an image
- Run
- 3961f27e
Measured on 2026-10-17: 128 runs finished, 9 models.See every measured run
Runs on your machines
One engine, wherever you work.
The engine
Products that run on Orazaka.
Orazaka is the engine behind Krizaka's own products. Each one says what it uses it for.
Cost & control
The same AI. Different ownership.
What changes when the AI runs at home.
Where do your prompts go?
- A cloud assistant
- To a provider's servers, under their terms.
- Orazaka
- To your machines. A cloud model only if you connect one.
What do you pay?
- A cloud assistant
- Per seat and per token, on their price list.
- Orazaka
- Your hardware, and credits you hold. A Studio shows its estimated cost before it runs.
Who decides?
- A cloud assistant
- The vendor: it changes models, limits and policies.
- Orazaka
- You do: which models run, who signs in, what each role may use.
What can agents touch?
- A cloud assistant
- What the integration was granted: its scopes decide.
- Orazaka
- What you approve. An agent shows its plan and acts once you approve it.
Questions
Before you ask.
Does anything leave my machines?
Models run on your hardware by default, and nothing phones home. A hosted model is used only if you connect one — and every result says where it was made.
What does it run on?
A Mac with Apple Silicon or a Linux machine, and your own servers for the platform. The measured runs name the exact machine, engine and model they were taken on.
What does it cost?
The software is open source (Apache-2.0). You pay for your hardware, and for a hosted model only if you connect one. A Studio shows its estimated cost before it runs.
Which models can it use?
Open-weight and hosted models for chat, vision, image, video and speech — the catalogue lists each one with its licence and where it runs.
Can I try it?
The public development site shows real recorded runs and the documentation. Accounts open when the development services run; until then, talk to us.
Bring AI in-house — not your data out.
Tell us what you want to run on your own machines; a person answers.