Connect your AI (BYOK)
Pyre is bring-your-own-key. It runs no AI of its own — you point it at a provider you choose, and your messages go straight from your device to that provider over HTTPS. No middle-man, no Pyre server reading over your shoulder. This is the one thing you must do before anything else works.
What you need
An API key from any OpenAI-compatible provider. That's a wide net:
- Aggregators that resell many models behind one key (OpenRouter, for example).
- First-party model APIs that speak the OpenAI format.
- Community proxies (the usual URL + key).
- A local server on your own machine or network — no key, no cost beyond the hardware you already own.
Free and cheap options exist
Several aggregators have free tiers, and a local server costs nothing to run. Pyre is free; what you spend is entirely between you and the provider you pick.
Add a provider
- Open More → API Connections.
- Add a provider and choose its kind — External, Proxy, or Localhost. It's only a UI grouping; they all speak the same protocol.
- Paste the provider's base URL and your API key. Pyre tidies the URL for you, so you won't end up with a stuttering
/v1/v1/. - Tap Browse models to pull the provider's model list, and pick one. (On a local server, the model you pick is usually loaded on demand.)
- Save. You're connected.
Keep as many providers configured as you like and switch the active one whenever the mood strikes. The AI Creator can even use a different, recommended model from the one you chat with.
Your key is handled carefully
Keys live in your operating system's secure store — not plaintext, and never in a casual backup unless you explicitly opt in. If a provider's error message ever leaks a token back at you, Pyre scrubs it before it hits your screen. See Your data & API keys.
Picking a model
- For chatting, most modern instruction-tuned chat models do nicely. Reasoning models work too — Pyre hides their
<think>scratchpad by default so you don't have to watch the model talk to itself. - For the AI Creator, the model matters more than the prompt. A capable, uncensored model (a DeepSeek-family model is a fine starting point) produces the most complete cards. In-app recommendations live under More → Character Creator.
If a provider fails or gets cold feet mid-scene and you have another configured, Pyre's smart fallback offers to retry on the next one. It never switches silently — your call, always.
See also
- Providers & connections — the full reference: provider kinds, extra params, context-window display, smart fallback.
- Your first roleplay
- Privacy by design