Skip to content

Glossary

Plain-English definitions of the terms used across these docs and inside Pyre, in alphabetical order.

Term What it means
Architect The Creator's model "persona" — the character/scenario architect you talk to while it builds a card. Its prompts ship as a forkable preset.
BYOK Bring-your-own-key. You supply the API key for a model provider; Pyre is just the interface and never hosts a model itself.
Branch / variant An alternate generation of a message. Swipe </> between variants; rewriting a user message keeps each branch's own downstream timeline so you can explore alternate paths without losing the original.
Character card (chara_card_v2) The standard, portable data model for a character — description, personality, scenario, first message, example dialogue, tags, and more — readable across the ecosystem and embeddable in a PNG.
Creator Pyre's conversational, self-healing card builder ("Creator v2"). You describe a character or world and the Sheet fills in as you go — no "Generate now" button.
LAN sync hub A desktop Pyre running a built-in LAN HTTP server (opt-in) that your other devices pair with to sync over your own network — no cloud, no account.
Live Sheet / Sheet The live chara_card_v2 data view in the Creator that fills in field-by-field as you talk to the architect, so you watch the card take shape.
Long-term memory A branch-aware auto-summarizer that condenses a chat into a running, continuous-narrative recap injected into the prompt, so old details survive past the model's context window.
Lorebook / World Info A named collection of trigger-keyword entries injected into the prompt when relevant (or always, if "constant"). Lets the model recall world facts only when they come up, saving context.
Macro / template token A placeholder like {{char}}, {{user}}, or {{scenario}} that Pyre resolves at send time so one prompt adapts to whatever character, persona, and lore are active.
OOC Out-of-character. An aside to the model that isn't part of the story — sent with the /ooc slash command.
Persona A reusable profile for you — name, description, avatar, voice samples, and bound lorebooks — so the model knows who you're playing.
Preset A named bundle of system-prompt template, post-history block, optional impersonate/continue overrides, and optional per-preset sampling — the whole "engine tuning" of a chat in one tap.
Provider An OpenAI-compatible endpoint you connect to (hosted service, community proxy, or self-hosted/local server) that actually runs the model.
Reasoning model A model (DeepSeek-R1 etc.) that emits a "thinking" scratchpad. Pyre understands it, wraps it in <think> tags, and hides it by default so your roleplay text stays clean.
Scenario / Narrator card A Creator mode that builds a whole setting — an omniscient narrator/Gamemaster that voices NPCs, with structured world and cast sections — rather than a single character.
Scene background The soft art behind the chat bubbles ("chat backdrop"), sourced from the character or persona avatar, a custom image, or none — with adjustable opacity.
Script Stage direction inside a chat — for example a scene-change narration sent with the /scene slash command — used to steer the roleplay without leaving the input box.
Smart fallback When a generation fails or is refused, Pyre offers (never silently) to retry on your next configured provider, and learns which providers tend to refuse.
Snapshot A frozen copy of the character(s) a chat uses, taken per chat, so editing a card later doesn't retroactively rewrite your old conversations.

See also