Ollama¶

The Ollama provider connects Missy to locally-running models via the Ollama REST API. No API key required -- everything runs on your hardware.

Setup¶

Step 1: Install Ollama¶

curl -fsSL https://ollama.com/install.sh | sh

Step 2: Pull a model¶

ollama pull llama3.2

Step 3: Configure Missy¶

providers:
  ollama:
    name: ollama
    model: "llama3.2"
    enabled: true

That is it. No API key, no network policy changes for the default local setup.

Step 4: Verify¶

missy providers
missy ask "Hello" --provider ollama

Configuration¶

providers:
  ollama:
    name: ollama
    model: "llama3.2"
    base_url: "http://localhost:11434"    # Default Ollama address
    timeout: 60                           # Local inference can be slower
    enabled: true

Remote Ollama server¶

If Ollama runs on a different machine:

providers:
  ollama:
    name: ollama
    model: "llama3.2"
    base_url: "http://10.0.0.50:11434"
    enabled: true

The base_url host is automatically added to provider_allowed_hosts, so network policy is handled for you.

Model tiers¶

providers:
  ollama:
    name: ollama
    model: "llama3.2"
    fast_model: "llama3.2:1b"
    premium_model: "llama3.2:70b"

Available models¶

Some popular models that work well with Missy:

Pull any model with:

ollama pull MODEL_NAME

How it works¶

Unlike the Anthropic and OpenAI providers, the Ollama provider does not use a vendor SDK. It communicates directly with the Ollama /api/chat endpoint via Missy's PolicyHTTPClient, which means all requests pass through the network policy engine.

The provider uses stream=false to receive the complete response in a single JSON payload.

Configuration reference¶