JARVIS UltraManualv1.31

How-to guides

Run agents on local models with Ollama

With Ollama on the same computer, agents can run on models that live on your machine: free, and nothing leaves the computer. A local model is usually slower and less capable than a hosted one, so it suits simple, private or high-volume work.

Before you start

  • Ollama installed and running on the computer that runs JARVIS Ultra.

  • At least one model that can call tools, pulled in a terminal, for example:

    ollama pull qwen3.5:2b
    
  • A 64,000-token context window. Ollama gives 4,096 unless you raise it. In the Ollama app, raise Context length to 64k in its Settings. On Linux, where Ollama runs as a service, set OLLAMA_CONTEXT_LENGTH=64000 for the ollama service (sudo systemctl edit ollama) and restart it.

Add Ollama

  1. Open Settings > Models and press Add provider.
  2. Pick Ollama on this computer ("Models you pulled with Ollama. No key, free.").
  3. Leave Address as it is, unless you changed OLLAMA_HOST. Only an address of this computer is accepted.
  4. Press Test and save. The card shows the number of models and "On this computer at 127.0.0.1:11434 · free".

If Ollama is not running, the message says how to start it on your system, for example "Open the Ollama app, then press Test and save again."

Give an agent a local model

  1. Open Team, click the agent and choose the Model tab.
  2. Pick the model under Primary AI model. Local models are listed under Ollama.
  3. Press Save. The button reads "Checking model with Ollama…" while Ollama loads the model once, which can take up to two minutes. The check refuses the model, and saves nothing, when Ollama does not answer, when the model cannot call tools, or when the window is smaller than 64,000 tokens. The message says which and how to fix it.

The same check runs on Hire and on Apply to all N agents. Turns on a local model may take up to 30 minutes and cost $0.00 in Settings > Usage.

Add or remove models later

Pull or remove models with ollama pull and ollama rm in a terminal, then press Test on the Ollama card. Test lists the models Ollama has now ("Ollama 0.40.16 answered. Models: …").