Docs

Agent Desktop/Work with your agent/Choose and configure models

Choose and configure models

Ask AI about Vinkius

Add local runtimes or online providers, configure a model and choose which model answers each chat.

Models is where you decide how your agent thinks. Agent Desktop supports local runtimes and online providers, so you can choose privacy, capability and cost per task.

Models

Connect providers, enable models, and choose the model agents use.

Stored securely
Models shows providers, local runtimes and a secure model configuration in one place.

Add a provider

Select Add provider and follow the three steps:

  1. Choose Provider: select a provider tile
  2. Configure Connection: set the display name, Base URL, API key when needed and API format
  3. Add Model: choose detected local models or enter a model ID manually

The provider gallery includes OpenRouter, Ollama, OpenAI, Anthropic, DeepSeek, Google Gemini, xAI, Z.ai, Qwen, Moonshot AI, Hugging Face and Custom.

Local models

For Ollama and compatible local runtimes, Agent Desktop can scan for available models. Select the models you want to enable. If no model is found, start the runtime and use Discover local models again.

Local providers use local endpoints. The app does not require a cloud API key for a local runtime.

Online providers

For an online provider, enter the provider URL and add a model ID. The API format can be Responses, Chat Completions or Messages, depending on the provider. API keys are stored securely by the app and are not shown to the chat surface.

A provider can exist before you add a model, but a chat needs at least one enabled model to respond.

Configure a model

Set the model ID, display name, context window and cost information when the provider asks for it. You can add, remove, enable or disable models later from the provider editor.

Select a model in chat

The model picker in the Canvas composer selects the model for the current chat. Changing the picker does not delete your conversation or move it to another Space.

Embeddings

The Models sidebar has a separate Embeddings entry. Use it when you want to configure workspace Memory search. Embedding search may use an authenticated Vinkius service even when the model that writes your answer is local.

Next steps