Skip to main content
Ainomos does not answer with a model of its own. For every question it collects the sources and rules that apply, sends them to a language model, and records which model answered. The model is replaceable: your rooms, rules, decisions and evidence records stay in Ainomos, whichever model you use.

Two jobs, two kinds of model

Ainomos uses models for two different jobs. An organisation can offer many chat models. It uses exactly one embedding model. These pages also call it the search model.

Platform models and your own models

Open Settings, then Models to see every model your organisation can use. Under each model’s name, the list says where it comes from:
  • Platform: provided by the platform. You can change its governance, such as its data classes or whether it is active. Its address and key stay with the platform.
  • Organisation: connected by your organisation, with your own key or your own endpoint. You can change everything about it.

On Ainomos cloud

Your organisation starts with chat models and an embedding model that the platform provides. There is nothing to set up before the first question.

On your own infrastructure

Nothing is preconfigured. You supply your own endpoints and keys. The installation needs at least one chat model and exactly one embedding model before it can answer a question. You can connect a provider such as OpenAI, Anthropic, Amazon Bedrock, Azure OpenAI or OpenRouter, or any model behind an OpenAI-compatible address, including models you host yourself. See Connect a model.

The starter credit

On Ainomos cloud, chat on the platform’s models is paid by a one-time starter credit.
  • The credit is for the whole organisation and for all platform models together.
  • It is used once. It is not topped up and it does not renew each month.
  • Embedding is not charged to the credit. Search and documents keep working after the credit is used up.
  • When the credit is used up, chat on the platform’s models stops. The chat shows a Starter credit used up card with a Connect a chat model button.
To continue, a platform admin connects a chat model of your own on the Models page. If your default chat model is still a platform model, the first chat model you connect becomes the default for every room. You can do this at any time, also before the credit runs out.

Default models

On the Models page, the section What each job uses names one model per job:
  • Answers questions: the default chat model. A conversation starts on it unless the person picks another model in the chat. Background work always uses it.
  • Embeds text for search: the embedding model. When the platform provides it, the page says so and you cannot change it.
In the chat, people pick from the chat models their room allows. See Asking questions.

Which models a room may use

Every room has a data class from C0 to C3. The class says how sensitive the room’s content is. See Data classes. Every model has a list of the data classes it may answer in, shown in the column May answer in. A room offers every active chat model whose list contains the room’s class.
  • Each class is allowed on its own. A model allowed in C2 is not automatically allowed in C1.
  • A room whose class no active model may answer in gets no answers at all. The Coverage by data class strip at the top of the Models page warns about this with Nobody can chat here.
  • The room’s Context panel in the chat names the model in use and how many models the room allows.

Where a model runs

Each model also says where it processes your data: An external model may answer in C0, C1 and C2. C3 is the sovereign class: an external model answers in C3 only as a recorded override. The connect form warns before you allow it, the edit form asks for a reason, and the audit log keeps the override. When a room uses an external model, the chat marks it Outside EU/CH and names the provider and the region.

Why switching models keeps your work

Your knowledge, rules, decisions, exceptions and evidence are stored in Ainomos, not in a model. For each question Ainomos hands the model what it needs and keeps the result. When you change the default chat model, connect another provider or retire a model, only the model that writes future answers changes. Rooms, rules and records stay as they are. Each evidence record names the model that answered, so older answers stay traceable.

Why the search model is locked

The embedding model is the one exception. Every stored passage was turned into a vector by one specific embedding model. Vectors from two different models cannot be compared, even when they have the same size. Once passages are stored, Ainomos refuses to change the embedding model, because it cannot yet embed all stored passages again. The Models page shows how many passages are stored. If you run your own installation, choose the embedding model before you import documents. Who can do this: everyone can pick an allowed chat model in the chat. Only platform admins manage models and change the defaults.