API key & providers

Zagens talks to LLM providers through your API key and ~/.zagens/config.toml (shared with the runtime sidecar).

First-time setup

1. Obtain a key from your provider (e.g. DeepSeek platform).
2. Open Settings → API Key in Zagens.
3. Paste the key — on desktop it is stored via the OS keyring when available.

The sidebar warns API key not configured until a valid key is present.

Vision bridge (optional)

Settings → API Key → Vision bridge configures the vision model endpoint (scans, describe_image, …):

• API key stored in the OS keyring (vision entry)
• Base URL / model written to [vision] in ~/.zagens/config.toml

Without it, read_office still works; scanned-document OCR paths rely on the vision bridge.

Multiple providers

You can configure several OpenAI-compatible providers and switch at runtime. Example endpoints: DeepSeek, NVIDIA NIM, Fireworks, OpenRouter, local vLLM/Ollama.

See config.example.toml in the repo for [providers] samples.

Model selection

Pick models in the composer or settings rail. Model IDs change with provider catalogs — treat docs as patterns, not guarantees.

Security

• Never commit keys to git or share exports containing credentials.
• Runtime token for the local sidecar is separate from provider keys.

Related: Vision tools · Tool approval · Model routing