Self-Hosting

The current repository does not include a separate Navvy backend service. Self-hosting means configuring the extension to call an OpenAI-compatible endpoint that you operate.

Why self-host the model endpoint

• Keep model traffic on infrastructure you control.
• Avoid sending browser-state snapshots to hosted providers.
• Test local models for internal workflows.
• Control logging, retention, and network access.

Local Ollama example

Ollama can expose a local OpenAI-compatible API. One common setup is:

ollama serve
ollama pull qwen2.5

Then configure a Navvy provider profile:

baseURL: http://localhost:11434/v1
model: qwen2.5
apiKey:

The empty API key is intentional for local Ollama setups that do not require bearer authentication.

OpenAI-compatible gateway

If you run an internal gateway, expose a chat completions route compatible with:

POST {baseURL}/chat/completions

Navvy sends a request body containing the selected model, temperature, messages, tools, and parallel_tool_calls: false. Tool results are parsed from the first tool call returned by the model.

Optional Docker Compose for a local model

The repository does not provide an official Docker Compose backend stack. A minimal local-model stack can run Ollama only:

services:
  ollama:
    image: ollama/ollama:latest
    ports:
      - "11434:11434"
    volumes:
      - ollama-data:/root/.ollama

volumes:
  ollama-data:

Start it with:

docker compose up

Then set baseURL to http://localhost:11434/v1.

Operational note

DOM snapshots can contain visible page text. If you operate a gateway, avoid retaining raw request bodies unless you have reviewed the privacy impact for your use case.