Web workbench
ZenStory Workbench
Write a novel with AI in the browser: outline, characters, chapters and versions in one project.
56 GitHub stars13 skills
What it is
ZenStory Workbench is the browser member of the family: a three-panel workspace (file tree, editor, AI chat) where outlines, manuscripts, characters and settings are files, and a multi-agent backend drafts, suggests and revises against them. It ships its own skills and an Agent API for Claude Code and OpenClaw, is self-hostable under MIT, and is available as a service at app.zenstory.ai with free and paid plans.
What you need
- LicenseMIT, open source.
- Where it runsA browser. The workbench runs at app.zenstory.ai.
- DocsThe workbench documentation: getting started, user guide, reference and troubleshooting.
How it works
- How do I start writing in the browser? Open the workbench, choose Novel, Short Story or Short Drama Script, type an idea or leave it blank, and the project is created; there is no agent to install. The quick start walks you through the first round.
- How do I turn an idea into my first scene? Separate approved facts from open decisions, save characters and a scene plan, then draft into a named file. The first-story walkthrough includes an original lost-property premise, file plan and three scoped requests, not a recorded model run.
- How do I revise without changing the whole story? Open the target file, name the references to read and quote the passage. Attach material items when relevant. The AI assistant guide explains context and continuation; the editor guide shows a local before/after revision. An instruction to suggest changes is not enforced read-only mode, and file operations do not all use per-change approval.
- Can I take the manuscript out of the workbench? The manuscript export guide explains combined TXT from saved draft/script files, chapter ordering and a separate review copy. It is not a backup of characters, materials, chat or history. Use the version guide to distinguish file recovery from project snapshots; keep important originals independently.
- Will AI Memory keep every character consistent? AI Memory is a four-field project summary, not proof that the model reads every file or a guarantee against drift. Keep confirmed settings, saved-scene progress and undecided ideas separate. The worked memory guide shows what to save, how context limits matter and how to correct a wrong summary without silently rewriting the manuscript.
- Must I split every chapter into many prompts? No fixed prompt count is required. You can request a complete chapter when its decisions, evidence and stopping point are clear; unresolved motives may need discussion first. The writing-handoff workflow follows one scene from accepted design through prose, local revision and a saved-state handoff.
- How do I turn reference analysis into my own scene? Locate a choice in the source, separate evidence from interpretation, then decide what changes in your own characters' actions and costs. The original material-analysis walkthrough moves from a recording passage to a different toy-car practice scene, not just renamed props. The materials introduction covers TXT limits, asynchronous analysis and explicit chat attachments; uploading a book does not put it all into every conversation.
What makes it different
- Saved files support version comparison and recovery, and the whole project can be snapshotted. New version entries depend on the save type, plan quota and coalescing of consecutive manual saves.
- Under the privacy policy, your writing is not used to train general models by default; AI processing is done by third-party model services, as the policy describes.
- ZenStory Workbench needs no agent skill pack, only a browser; self-hosting means configuring the backend and model services yourself.
Who it is for
- Novelists who want a workspace rather than a terminal
- Writers managing long projects with version history
Compare writing workflows
Sources
Based on the source as of 2026-10-15; links point at that version.
- Agent API, file editing and version features
- Self-hosting prerequisites and model configuration
- Published policy on providers and general model training
- Default subscription limits include file versions
- Empty-idea quick creation
- Focused file, attachments and quotes
- Exported file types and ordering
- Project memory is budget-bound context
- Star counts are a 2026-09-29 snapshot; the workflow notes below reflect this source review. 2026-10-15: user-version quota and save coalescing · Plan quota for user versions