Agent · AI workbench

An agent that knows your shelf
and shows its work

The Agent tab hosts an AI agent that does real jobs: it reads your novels, completes worldbooks, organizes reading notes, and starts roleplay sessions from character cards. A native tool-call loop actually reads and writes files; reasoning and every tool step render in the open. Type @ to jump between books and cards; the skill catalog is yours to edit.

Get it on Google Play See roleplay

Native tool calls · read / write / edit / exec
@ references novels and character cards
Visible runtime · stoppable anytime

Phone · Agent workbench

9:41

Agent

What should the agent do first?Pick a common task — reference material first, then edit the drafted request.

Long-fiction readers and card players are missing a pair of hands

Lore falls behind the chaptersThe deeper the book, the larger the worldbook debt. Ask the agent to read the latest chapters and file new characters and places into the worldbook — it opens the file and writes, not just talks.
Notes stay in “later”After a night of reading nobody organizes notes. One sentence and the agent distills tonight's chapters into a page in your workspace, bookmarked for tomorrow.
Switching modules breaks the flowMid-chapter and want a session with some card? Type @ and the name — the agent opens the card and starts the session on the route you asked for, without leaving your reading.
Black-box AI you can't auditEvery reasoning pass, file read/write, and command renders as a card. Leave and come back — running tasks restore in place. Stop anytime; partial output is kept.

Skills are files. Edit the file, change the agent.

The agent's abilities hang off a skill catalog: each skill is a SKILL.md in the workspace describing what to do and what to produce. Card authoring, sticker packs, and theme building ship built-in — or write your own; change one line and the behavior changes. Models are swappable too: DeepSeek, Claude, Gemini, GPT, on your own keys.

FAQ

Common questions

What can the agent do with my books and cards?

It works inside a sandboxed workspace: reading novel chapters, character cards, and worldbooks; writing notes, new worldbook entries, and session openings; and running restricted commands for batch chores. Every read and write renders as an expandable tool card.

Will it rewrite my novel text?

No. Novel text is read-only to the agent; its output goes into companion data — worldbooks, notes, sessions. Every file write is logged on a tool card and can be rolled back. Stop a task anytime and generated content is kept.

What are @ references?

Typing @ in the composer lists your novels and character cards. Pick one and the agent can open that book or card directly — “start a rain-night session with @Luoxi” is one sentence, one cross-module jump.

Which models run the agent?

Any model with tool-calling: DeepSeek V4, Claude, Gemini, and GPT families are field-tested, on your own BYOK keys. Swap models from the composer; reasoning effort is adjustable.

Now on Google Play

Hand the chores to the agent. Keep the reading and the writing.

The Android app is live on Google Play; for the iOS beta we send one short note when the next wave opens. You can also email [email protected].

Get it on Google Play →

An agent that knows your shelfand shows its work