How to roleplay with Claude Set the story up once, then play it scene by scene.

Give each story its own project

Claude can narrate a whole world or speak as a single character, and the way you set it up decides how well a long story holds together. Start with a project. Anthropic’s help center describes projects as self-contained workspaces, each with its own chats and knowledge base. Every Claude account can make them, and the free plan allows up to 5.

A project gives a story what a loose chat lacks:

• Project instructions, which shape Claude’s replies in every chat inside the project. The world’s rules go here, so you write them once.
• Project knowledge, where you can add documents and text. Your story summary and character notes live here.
• A memory of its own. Inside a project, Claude uses that project’s memory plus your general memory, never another project’s. To seal a story off completely, switch off Use account memory on the project’s page.

Keep to one story per project. A second campaign gets a second project, so its notes and memory never mix with the first.

Write the rules Claude reads every time

Project instructions work best when they read like a brief to a game master: who Claude plays, where the story is set, who you are, and the handful of rules that matter most. Anthropic’s advice for keeping Claude in character points the same way: give a character a detailed personality, background, and quirks, and describe the situations you expect.

Paste this into the project instructions, then change the details:

Keep the instructions to the essentials. Long backstory belongs in project knowledge, where you can change it without touching the rules. Stories follow the same Usage Policy as any other Claude chat.

Keep your character in your hands

The habit to watch for is Claude writing your part: a reply in which your character suddenly agrees, flinches, or leaves the room. The rule in the instructions does most of the work. Back it up in play:

• Fix the first slip. Each reply sets an example for the next, so edit your last message and send it again, or reply with a short note: ((You wrote Ilse’s answer for her. Rewrite the reply and stop before she speaks.))
• Say where replies should stop. ((End each reply as soon as someone speaks to Ilse.))
• Leave your character out of the examples. An opening or sample dialogue in which your character speaks reads to Claude as permission.
• Describe your character in one place. A short block in the instructions, or a file in project knowledge, so Claude isn’t guessing.

A character block that gives Claude enough to work with:

Open the scene and take turns

Your first message sets the scene and hands Claude the first move. Give it a place, something already going wrong, and a reason for your character to act:

After that, each turn says what your character tries or says, and the outcome is Claude’s to describe. Turns that give Claude something to answer:

• Ilse pulls the door shut, counts to ten, and opens it again. “Tickets, please.”
• She asks the old man in seat 14 whether he heard the door, watching his hands instead of his face.
• She tries her brass key in the lock of carriage 9, half hoping it won’t turn.
• ((Skip to the next station. Nothing happens on the way.))
• ((Slow down. Let this conversation take its time.))
• ((Bring in a passenger who knows more than they should.))

If a reply misses, ask for another version or edit your message instead of arguing inside the story, where the argument becomes part of the plot.

What happens as a chat grows long

Every Claude chat has a context window: the amount of conversation Claude can work with at once. Anthropic’s help center explains what happens on paid plans with code execution turned on: when a chat nears the limit, Claude summarizes earlier messages so the conversation can carry on, and the full history stays available for it to refer back to. On any plan, a new chat starts with a fresh context window.

For a story, two things follow:

• Early scenes move into a summary. The full history stays available for Claude to look back at, but small details, such as a name, a promise, or the number on a door, are easy to miss once they’re summarized.
• Long chats use more of your allowance. Anthropic notes that conversations long enough to be summarized consume more of your usage limit.

Claude’s memory helps around the edges. It’s on by default on the Free, Pro, and Max plans, saves what it learns as a set of topics while you chat, and lets you review and edit those topics in its settings. You can also tell it in a chat what to remember or forget. Memory keeps topics rather than scenes, so keep your own summary as the story’s record.

A summary routine for long stories

The fix that works in any chatbot is to keep the story’s memory yourself and hand it over whenever a new chat starts. In Claude, a project makes this tidy:

1. Recap at the end of each sitting. Ask for it in a fixed shape, so each version is easy to compare with the last.
2. Save it to project knowledge. Keep one document called Story so far, and replace it each time instead of piling up versions.
3. Start a new chat when details slip. Open it in the same project and paste the resume prompt below.
4. Check the first reply. If Claude gets something wrong, correct it once, then fix the summary so the mistake doesn’t come back.

The recap prompt:

And the prompt that opens each new chat:

Why long stories move to LoreWeaver

Claude is an excellent general assistant, and a project with a good summary can carry a story a long way. The routine above is still bookkeeping you do by hand, every sitting, and months-long stories are where it piles up. LoreWeaver is built so that work happens on its own: a storyteller runs the world, and the story’s record builds up while you play.

• Memory written for you. Every 4 replies, LoreWeaver saves what happened as memories, and the ones that fit the moment come back by meaning, wording, and who is in the scene. A running recap covers the whole story. You can edit the recap, and pin or add memories of your own.
• Lore cards. Each person, place, or thing gets a card that joins a reply when the story names it, or every reply once pinned. On Plus and up, Lorekeeper suggests cards for the people and places the story invents, and the LoreMap shows how they connect.
• The Loom. Story threads fire planned events when their conditions are met, and trackers keep health, quests, inventory, and dice checks on screen.
• Claude, with all of that built in. LoreWeaver runs Claude Haiku 5.5 itself, so you need no Claude subscription or API key. It opens on Plus, and free accounts can play it with their 25 premium replies a day. GPT and Gemini models are there too, among 43 models in all, and you can change models mid-story without losing its memories or lore.

Any community world can be played as a guest for 20 replies, with no account. After that, the free plan keeps going with unlimited replies from a fast free storyteller, about 5 to 7 seconds for a typical reply, plus 25 premium replies a day. LoreWeaver accounts are for adults, 18 and over.

Bring a Claude story into LoreWeaver

Your summary and character notes are all it takes to carry a story across:

1. Build the world. In the world builder, describe the story in a sentence or two. It writes the premise, the people and places as lore cards, who you play, and an opening, and you can change any part before you begin.
2. Save your character as a persona. Paste the character block, then apply the persona to the story from the Character drawer.
3. Paste your summary into the recap. The story’s recap, in the Memory panel, is read before every reply, and later updates build on what you wrote.
4. Move your table rules into session notes. The storyteller already avoids speaking or deciding for your character, so keep notes for rules of your own, like “keep the supernatural uncertain until Vossburg”. It reads them on every reply, up to 500 tokens.
5. Choose Claude Haiku 5.5 and play on. From here, memories and the recap build up on their own.

Sources

• Claude Help Center: What are projects?
• Claude Help Center: Use Claude’s chat search and memory to build on previous context
• Claude Help Center: How do usage and length limits work?
• Claude Docs: Increase output consistency, on keeping Claude in character
• Anthropic Usage Policy

All checked October 30, 2026.