Concepts

Orazaka is an AI platform you run yourself. Your documents, prompts and results stay on your machine. There is no telemetry.

Three layers

1. Clients. The web app, the mobile app and the orazaka command line. They never talk to a model directly.
2. The engine. One Java backend. It checks who you are, routes each request, and runs it through the pipeline.
3. Workers and models. Local models (through Ollama) answer chat. Workers do the heavy jobs, such as images and video, and report back.

A request, step by step

You ask for something. The engine does five things, in this order:

1. It knows who you are and what your plan allows.
2. For a Studio run, it checks the content policy before any cost.
3. It enriches the prompt with context and picks a model.
4. It runs the job, on your hardware or on a provider you chose.
5. It writes a run record: what ran, where, how long, what it cost.

The ordered filters in steps 2 and 3 are the interceptor pipeline. A fixed core chain runs first, then the interceptors an operator configured in the database.

Packs and Studios

• A Studio is a workflow: a few steps (model calls, transforms, human approvals) behind one form.
• A pack is a folder that ships Studios. Installing a pack adds data, not code.
• A run is one execution of a Studio. Every run has an id, a price and a result you can open again.

Jobs and workers

Anything that takes more than a moment is a job. Jobs travel over a message queue to a worker, which can be written in any language. The queue contract is in the worker protocol.

Who decides what

• You, the operator, decide the models, the plans, the content policy and who may sign in.
• Nothing is refused by default on a fresh install, except content that sexualises minors. That one is always refused.

Where the details live

The reference pages are generated from the code, so they are always current: the API, the CLI, the models, the architecture with its modules, interceptors, business layer and decision records.

Next

• Quickstart: run it in five minutes.
• Run a Studio: see a run end to end.
• Content policy: what you can allow or refuse.