Providers & Models

Apis AI talks to local and external LLMs through a unified provider model. Everything is managed in the Settings, then Providers & Models tab.

Provider Types

Every provider has a type that determines the request format:

Predefined providers set the matching type automatically – for a custom provider you choose it yourself.

Creating a Provider

1. Choose a provider – in the Add provider area, select either a predefined provider or Custom provider.
2. Enter the Base URL – the endpoint URL (e.g. http://localhost:11434/v1 for Ollama or https://api.deepseek.com).
3. Add an API key – often not needed for local models, required for cloud providers.
4. Add – the provider appears as a card and loads its model list.

The Fields in Detail

Via Advanced settings you reach additional options such as the max tokens field and the thinking toggle.

Models, Favorites & Tool Support

• Model list – filterable by search. For long lists a search field with a scrollable selection appears.
• Favorites – mark frequently used models with the star; they appear at the top.
• Tool-call support – models without tool-call support are marked accordingly. Such models are not suitable for MCP or file tools.
• Tool selection per model – when a model's context window is too small for all tool definitions, switch off individual MCP servers and tool groups for exactly this model – see Settings, Tool Selection per Model.
• Detect context window – queries the appropriate token limit from the provider.

Where the Model List Comes From

Apis AI does not ship a fixed model catalog. For OpenAI-compatible providers (DeepSeek, OpenRouter, Ollama, Groq, …) and for Gemini, the app fetches the list live from the provider's /models endpoint whenever the provider loads. Claude uses a built-in list. The selection therefore shows exactly what the provider's API reports.

Manual models (local providers only). For local providers (e.g. Ollama or your own local endpoint) you can enter models yourself under Manual models. That list then replaces the API list – it does not add to it; clear the field to show the provider's own list again. For cloud providers such as DeepSeek this field is not shown, so their selection stays exactly the API list.

Local Llama Server

If an OpenAI-compatible provider points at a local llama.cpp server (a base URL on localhost:8080, for example http://localhost:8080/v1), Apis AI starts that server for you as soon as the model list is loaded and nothing answers yet. It uses the llama / llama-server binary from your PATH together with the model you configured.

Apis AI stops the server again when you close the app – but only the one it started itself. That way a local model cannot keep your graphics card's memory occupied after Apis AI has been closed. A server that you started yourself, or one that was already running, is left untouched.

The selected model is really loaded, the server restarts when you switch models, and you can give each model its own server arguments (flags) and presets. Details are on the dedicated page Local Llama Server.

Default Provider & Model

Under Default provider & model you define what new chats and workflow steps start with. The default is also shown in the sidebar. In individual chats and workflow nodes you can deviate at any time.

Thinking & Max Tokens Field

Thinking (Extended Thinking)

The thinking toggle makes the model "think" before answering. It is configurable per provider and can be overridden per conversation via the Thinking row in the chat's "+" menu. The provider value serves as the default; the conversation value takes precedence.

Max Tokens Field

Determines how the maximum output length is transmitted to the model. The Automatic setting detects the appropriate field itself – that is the right choice for most providers.