Choosing an AI Backend

DGAI can talk to a cloud AI (most capable; needs an API key) or a local AI running on your own machine (fully private, no key, but you need a capable computer). Pick from the backend dropdown at the top of the window.

Cloud providers

Cloud models are the most capable. With any cloud backend, your data is always tokenized — real identities never leave your machine (see How Your Privacy Works).

Entering your key

Open Privacy & Debug → Settings, paste the key into the {Provider} API key field, and press Open again. Alternatively set the provider's environment variable (e.g. ANTHROPIC_API_KEY), which takes precedence over a stored key. Keys are stored locally in DGAI's settings file.

Local AI (nothing leaves your machine)

Run a model on your own computer — no API key, no cloud:

• Ollama (http://localhost:11434) — default model llama3.1:8b. Use a tool-capable model (e.g. llama3.1, qwen2.5).
• LM Studio (http://localhost:1234) — runs any model you've loaded.
• Custom — any OpenAI-compatible endpoint you point it at.

Real-names mode (local only)

With a local AI, you can optionally turn on Real names (localhost AI only) in Settings — DGAI then skips tokenization and the model sees actual names. Because the endpoint is on your own machine, nothing leaves your computer regardless. DGAI refuses this mode for any non-localhost endpoint.