How chatting works
A conversation on Lyra is a series of turns — you write, the cast replies. The model stays in character and keeps the thread of the scene, and a handful of controls let you shape every reply.
Turns and the cast
Each message you send and each response from the cast is a turn. You write as yourself or as a persona; the characters answer in voice. The cast stays grounded in the story’s world — its setting, the characters, and what is happening now — so replies stay in character rather than drifting into generic chat.
Memory
The cast remembers the scene as it unfolds — earlier moments, your choices, and the details that matter carry forward, so the story stays consistent instead of resetting. Higher plans let the cast hold more of the story in mind at once, which keeps longer arcs and callbacks on the thread. Lyra+ adds auto memory on top: durable notes the storyteller keeps about you and recalls later. See Plans for what each tier includes.
Shaping a reply
Hover or tap a reply from the cast to reveal its actions:
Under each reply: the media icon (image, comic, video), swipe between alternatives, Regenerate and Edit. Branch, Rewind and Delete are in the reply's menu: ⋯ on a computer, press and hold the reply on a phone.
·
Regenerate — not feeling a response? Generate a fresh take on the latest reply.
·
Swipe between alternatives — when a turn has more than one version, page through them and keep the one you like.
·
Edit — fine-tune the wording of a reply to keep the scene exactly on track.
·
Branch — fork the conversation from any point to explore a different path without losing the original (in the reply's menu).
·
Delete — remove the most recent turn to back up a step (in the reply's menu).
·
Rate — give a reply a thumbs up or down to signal what worked.
Branching creates a separate copy of the chat up to the point you chose, so your original storyline stays intact — handy for trying a bold choice without committing.
Tune the chat
Tap the model’s name in the message bar to switch the model. Under ＋ › Settings, choose a persona to write as, add free-form guidelines for tone and content, and turn on reasoning when you want the model to think through a complex scene before it answers. Each is covered in its own guide.
