Skip to content

Connect your AI (BYOK)

Pyre runs no model of its own. You choose a provider, and Pyre becomes the roleplay client around it.

Native apps

On Android and desktop, Pyre can call providers directly. Add a connection under More -> API Connections:

  1. Choose the provider format/kind.
  2. Enter the endpoint URL.
  3. Enter the API key, unless your local server does not require one.
  4. Browse or type the model id.
  5. Test the connection and save.

Supported paths include OpenAI-compatible endpoints, Anthropic-native API, hosted aggregators, community proxies, and local/LAN model servers.

Browser/PWA

A browser client paired to a desktop or self-host hub does not need a local provider key. The browser streams chat to the hub, and the hub calls the provider with its own securely stored key.

That means a school/work/borrowed browser can use your synced Pyre library without receiving your secrets.

Picking a model

  • Roleplay chat usually wants a strong instruction-tuned model that follows character context well.
  • Creator benefits from models that follow structured JSON and do not collapse under long card-building prompts.
  • Vision needs a vision-capable model if you attach image references.
  • Localhost models may need longer warm-up; Pyre gives local providers more patient timeouts.

If a provider fails or refuses a turn and you have another configured, smart fallback can offer to retry on the next one. It always asks first.

Key safety

Native app keys live in OS-secure storage. Normal backups exclude keys. If you include the Connections category in a backup, Pyre warns because that backup file will contain secrets.

Browser clients never store provider keys.

See also