Skip to content
Browse documentation

Connections & models

Set up Ollama and local models

Load a locally installed model, set the endpoint, and distinguish discovery from inference.

Prerequisites: install Ollama using its current supported distribution, start its local service, and choose an actual model from its catalogue appropriate to your computer. Vopit does not install or download models for you. Local inference still requires sufficient RAM, storage, and any relevant GPU support.

Prepare a real model locally

Local terminal — replace the model placeholder with an Ollama catalogue name
ollama --version
ollama pull <model-name>
ollama list
ollama run <model-name>

Check that the selected model appears in ollama list and can answer a short prompt in ollama run. If the service is not already running, start it with ollama serve in a separate terminal; do not launch a second listener on the same port.

Connect Vopit to Ollama

  1. Open Settings → Connections → AI connections → Local models → Configure Ollama.
  2. Use Server endpoint http://127.0.0.1:11434/v1 for the default local service. HTTP is allowed only for loopback endpoints; use an HTTPS endpoint for a remote compatible service.
  3. Choose Save & load models. Set Connection default model to the exact model ID discovered locally. Ollama has no selected model by default. Optionally enter a smaller local model as Fast model for the Ultra-fast and Balanced speed modes; left empty, they use the connection model.
  4. Choose Use as default and Save settings. Check How your agents run for overrides, then run a small read-only request and inspect the response.
  5. If discovery fails, confirm the listener, endpoint, and model installation. If inference fails, test the same model in the local terminal and inspect resource availability. Correct the cause before resuming a request.

Correct guessed model capabilities

Vopit infers whether a model accepts a temperature or reasoning parameters from its name. Local model names often do not follow those conventions. If requests fail on an unsupported parameter, expand Request budgets and model capabilities in the Ollama connection and set Temperature support (auto, supported or unsupported) and Reasoning capability (auto, enabled or disabled) explicitly. The same panel sets the Response token budget and the Model catalogue timeout (ms) and Connection test timeout (ms), which can help with a slow local server. Restore connection defaults returns to automatic detection. For Ollama, Advanced: endpoint & cost estimates contains only optional cost rates; the endpoint is the Server endpoint field.

All documentation