FAQ
Short answers to the questions people ask most. Each one is grounded in how the app actually works — no marketing fog.
What is Pyre?
A bring-your-own-key (BYOK), local-first AI roleplay chat client for Android, desktop (Windows / Linux / macOS), and the web. You supply a model provider; Pyre is the interface — characters, group chats, branching, an AI card creator, lorebooks, presets, and long-term memory. No Pyre backend, no account, no cloud. Made by Ember Team.
Is Pyre free?
The app is free, obviously. Pyre is BYOK — it talks directly to whatever model provider you configure, so you pay that provider for usage (or run a local model, which costs nothing to run). Pyre doesn't bill you and takes no cut.
Do I need an API key?
Yes. Pyre does nothing until you point it at a provider. Add an endpoint and a key once under More → API Connections, and you're chatting. A self-hosted/local server is the exception — most local servers ignore the key, so leave it blank there.
Does Pyre see my chats? Is there a Pyre server?
No. There is no Pyre backend and no account. Every request goes device → your provider over HTTPS, no middle-man — Pyre never proxies your chats. Your characters, chats, presets, lorebooks, and personas live in app-private storage on your device, where the ban button can't reach.
Info
The one optional exception is the LAN sync hub, which runs on your own desktop on your own network — never on a Pyre server. See Cross-device sync.
Is my content moderated or filtered?
Pyre hosts, generates, and moderates nothing. There's no content layer between you and the model — no nanny standing in the doorway. What you write and what your model says is between you and the provider you chose. Pyre is for adults (18+).
Which providers work?
Any OpenAI-compatible endpoint — anything that speaks the standard chat-completions protocol. Hosted services (OpenRouter, Venice, NanoGPT, for example), community proxies, and self-hosted/local servers all qualify. Keep several configured and switch the active one whenever. See Providers & connections.
Which model should I use for the Creator?
For card-building, the model matters more than the prompt. A DeepSeek-family or other uncensored model tends to give the best results — a model that flinches at its own shadow makes thin cards. Pyre shows a recommended-models banner in the Creator so you don't burn time on a bad fit — find it under More → Character Creator.
Can I import my existing cards, presets, and lorebooks?
Yes.
- Cards —
chara_card_v2PNGs (the community-standard Tavern format, v1 and v2) and raw card JSON, plus import-by-URL. - Presets — the importer parses the common community prompt-pipeline format and maps its markers into Pyre's template tokens, carrying over sampling and impersonate/continue overrides.
- Lorebooks — standalone import/export, plus books embedded in a card.
Can I use Pyre on multiple devices?
Yes, over your own network — no cloud account involved. A desktop Pyre can act as a LAN sync hub: pair your phone by scanning a QR code, and your library (characters, personas, chats, presets, lorebooks) follows you. A browser can even run the web app through the desktop so your API key never leaves it. See Cross-device sync.
Is Pyre open source?
It's intended for release as open source; Ember Team is finalizing on AGPL-3.0. Pyre is currently in public beta — stable enough for daily use, with the rough edges a beta exists to find.
Is there an iOS build?
Current targets are Android, desktop (Windows / Linux / macOS), and web/PWA. Android is the primary mobile platform; on desktop, Windows is the most complete (some features fall back to the external browser on Linux/macOS).