Model Routing: decide which AI model handles which tasks, and when
ZiyaOS is not tied to any model. You connect cloud models and local models and decide which security level may use which model. Or you stay completely local.
TaskSummarize a personnel fileStrictly confidential
Small model, EU
Large model, EU
Local model
How it works
1
Task
An agent, an app or a workflow starts a task. It carries a security level.
2
Check the level
ZiyaOS looks up which models are allowed for this level and this connection.
3
Pick the model
The task goes to an allowed model: a small one for simple work, the local one for confidential work.
4
Log
The log shows which model saw which data.
Every request goes to the right model.
ZiyaOS picks the model per request by task, security level and cost. Simple work goes to small, cheap models, confidential work to the local model, only hard work to the large ones.
Virtual model
Small model
Medium model
Large model
Local model
If a provider goes down, your AI keeps working.
If a model provider goes down or hits its limit, ZiyaOS routes requests to another one automatically. Your agents and apps never notice.
Model providers
Azure OpenAIonlinedisrupted
Amazon Bedrockonlinedisrupted
Google Vertexonlinedisrupted
Through ZiyaOSonline
Model Routing in action
In these solutions every task goes to the model it is allowed to use.
Run local AI
Every task goes to the model its security level allows.
Cloud models from the EU or open-source models on your server. Switch without rebuilding when a better one arrives.
1
2
3
Security levels per task
From public to strictly confidential. Every level has its allowed models.
Model locks per connection
The HR database only goes to the local model. No matter which agent asks.
€
€€€
€
Cost per model
Simple tasks go to small models. You see what each model costs.
Frequently asked questions
Which AI models can I use?
You can use all common AI models, through OpenAI, Anthropic, Google, AWS Bedrock, Microsoft Foundry and OpenRouter. You can also connect your own models, including ones running locally on your premises.
Can we use our own models or our Azure OpenAI instance?
Yes, you can use your own models and your Azure OpenAI instance. We call it Bring your own AI. You keep your contracts with your existing providers and get cost control on top.
Can ZiyaOS run completely locally?
Yes, ZiyaOS can run completely locally: on-premise with local models, even completely offline. Local models at the level of the top models quickly cost six figures in hardware today. Local models are enough for simple tasks, often not for demanding ones. Also check where your data already lives: if you use SharePoint or M365, it is already at Microsoft. The pragmatic route is usually this: you bring models hosted in the EU, and ZiyaOS controls who may work with which models.