The considered workspace for long manuscripts.
Wendell holds the validated assembly, the provenance, and the rubric — so the page in front of you is the one you signed for.
Every concept carries a derivation chain. Six months later, you can still tell your idea from a model's.
Human-in-the-loop.
LLM output arrives as proposals. You accept, reject, or modify — explicitly.
Quantitative readiness.
Per-scene rubric scores.
Most tools collapse research, plan, and prose into one opaque project file.
You end up juggling three different things at once:
- Inventory — hundreds of source files, notes, interviews, drafts
- Assembly — the reading order: which scene goes where, what foreshadows what
- Production — the final format: DOCX, FDX, LaTeX, ePub
When these collapse into a single binder or vault, there's no diffable boundary between research, plan, and prose. AI assistants make this worse, not better — they write directly into your draft with no record of where ideas came from.
Wendell separates the three. The plot graph is a validated, queryable contract. The corpus is a tracked inventory. Production formats compile out of the assembly, not into it.
Four things no other tool combines.
Each is available somewhere. Wendell is the only place all four ship together.
01
Provenance-first assembly
Every concept in your assembly carries a traceable derivation chain — source file, LLM output, or prior draft. Provenance is structural, not a tag you remember to add.
02
Registry-backed relations
Cross-scene relations are typed: foreshadows, resolves, contradicts, rhymes_with, shares_motif. Queryable in aggregate. Backed by a graph, not free-text labels.
03
Human-in-the-loop AI
LLM output returns as Accept-route proposals. Nothing touches your canon without explicit signature. Every accepted change writes a ledger event and updates the provenance chain.
04
Quantitative readiness
A per-scene rubric produces completeness scores. Outcome scorecards and narrative maturity bundles make a 300-scene corpus triageable. Quality assessment stops being subjective.
The most important distinction is what happens after the AI responds.
Black-box AI writes to your work directly. Wendell's HITL gate means every model output is a proposal until you sign it.
| Black-box AI (most tools) | HITL assembly (Wendell) |
|---|---|
| Obsidian plugins write to notes directly | LLM returns proposals as review cards |
| NotebookLM answers in a chat that doesn’t mutate anything downstream | Each proposal is Accept / Reject / Modify |
| Notion AI rewrites blocks in place | Only on Accept does the assembly mutate |
| Pasted Scrivener output enters the binder with no derivation | Every accepted change writes a ledger event |
| You audit afterward. Drift accumulates silently. | Nothing reaches canon without your signature. Drift is impossible to hide. |
This is not a workflow preference. It's a trust architecture.
Who it's for
The Complex-Work Author
You're running a multi-POV, multi-subplot project. 200+ scenes, 500+ corpus files. You've outgrown Scrivener's binder for cross-scene causality. You're comfortable with git and Markdown.
The Author-Researcher
You're an academic, longform journalist, or thesis writer. You need provenance discipline applied to sources, interviews, and drafts. The corpus and ledger are what you're here for.
The Small IP Room
You're a game studio, transmedia team, or production company. "Our Scrivener project" is not enough when a deal asks for change history and rubric evidence.
How it works
Three layers. Each one owned by you.
Inventory.
Drop your corpus into the repo. Sources, interviews, prior drafts. Each file gets a hash; nothing changes silently.
Assemble.
Build the plot graph with typed edges. Score scenes against a rubric you own. Watch the maturity bundle update as you work.
Compile.
Export to DOCX, FDX, LaTeX, or ePub when you lock for production. Your assembly is the source of truth. The output is downstream.
Pricing
Pricing will be announced at general availability. Join the waitlist to get founding-member pricing.
Frequently asked questions
No. Scrivener is a prose surface. Wendell is the truth layer underneath. Keep writing in Scrivener (or anywhere else); let Wendell hold the validated assembly, the provenance, and the rubric. Compile out of Wendell when you ship.
No. Your assembly, corpus, and ledger never leave your control without an explicit action you take. See the Privacy Policy for the full architecture.
Obsidian proved authors will run a local-first knowledge tool. Wendell adds the structured contract Obsidian leaves to plugin discipline, and the AI gate Obsidian's plugin authors decline to enforce. The two can coexist; many of our users will already have an Obsidian vault.
Because revision is the job, and git is the most-tested versioning system on earth. Your manuscript deserves the same audit trail your code has.
Waitlist members are notified first. Sign up below.