Help Center
Everything you need to use Cherry Edits. There are two things the platform does: edit a finished manuscript (the four-stage editing pipeline) and help you write new chapters (Assist). Pick the workflow that fits where you are in your book.
Getting Started
Three steps to your first AI-powered edit.
Create an account
Click Sign Up in the top-right corner. Enter your name, email, and a password. You'll be logged in immediately—no email verification needed.
Add your API key
Go to Settings (user menu, top-right). Choose Anthropic or OpenRouter as your provider, then paste your API key. Cherry Edits uses your key to call the AI directly—no markup, no middleman.
Create your first project
From the dashboard, click New Project. Choose a scope (Standalone for a single book, or Series for multi-book continuity). Give it a title, pick a genre, and you're off.
Tip: Don't have an API key yet? You can get one from Anthropic (console.anthropic.com) or OpenRouter (openrouter.ai) in minutes. Both offer pay-as-you-go pricing.
Standalone Editing
The full workflow for editing a single manuscript, from upload to download.
1. Upload your manuscript
Drag and drop a .docx, .txt, or .md file. Chapters are detected automatically from headings, "Chapter N" patterns, or common separators. You'll see the detected chapters before anything runs.
2. Configure your edit
Select which editing stages to run (pre-edit, developmental, line, copy). The page shows a cost estimate and time estimate before you start. Toggle beat detection to add story structure analysis. Adjust batch size if you want smaller processing groups.
3. Watch it run
The progress page shows real-time updates as each chapter is processed. You can leave and come back—progress is saved per chapter. If the pipeline stops for any reason, it resumes from where it left off when you re-run.
4. Download results
Download the final edited manuscript, individual per-stage outputs, and the story structure report (if beat detection was enabled). You can re-run the pipeline anytime with different settings.
Chapter formatting tips
- Word (.docx): Use Heading 1 or Heading 2 styles for chapter titles—this gives the most reliable detection.
- Plain text (.txt): Use "Chapter 1", "CHAPTER ONE", or numbered headings like "1." at the start of a line.
- Markdown (.md): Use
# Chapter Titleor## Chapter Titleheadings. - Avoid putting chapter titles in the middle of paragraphs—they should be on their own line.
Series Editing
Edit multiple books with cross-book continuity awareness. The platform tracks characters, timelines, and world-building across your entire series.
Create a series project
Click New Project, choose Edit mode and Series scope. Give the series a title and select the genre.
Upload your books
Add two or more books. For each book, provide a title and upload the manuscript file. The books are listed in order—you can rearrange them to match the series reading order.
Run continuity analysis
The AI reads every book and extracts characters, timelines, world-building details, and plot threads. Then it cross-references everything to find inconsistencies. This produces a series bible and a continuity report with severity-ranked issues.
Review continuity results
Browse the continuity findings by severity (critical, major, minor). Review character profiles, timeline events, and world rules in the series bible. Download the full report for reference.
Resolve issues (optional)
If critical or major inconsistencies are found, a Resolve Issues button appears. This opens a resolution wizard where you review each issue, choose a canonical value (e.g., the correct character name), and approve or skip individual fixes. Approved resolutions are applied to your manuscripts before editing begins—so the editing pipeline works with clean, consistent text. You can skip this step entirely and go straight to editing if you prefer.
Edit books one at a time
Each book gets the full editing pipeline (pre-edit, dev, line, copy) with the series bible injected as context. The AI knows your characters, rules, and timeline as it edits. If you resolved issues in the previous step, the pipeline uses the corrected manuscripts. After one book finishes, click Edit Next to move to the next one.
What continuity analysis checks:
- Character consistency — names, appearances, abilities, relationships, backstory
- Timeline conflicts — events that contradict each other or happen in impossible order
- World-building contradictions — rules, geography, magic systems, technology
- Plot thread continuity — unresolved threads, dropped subplots, inconsistent motivations
How resolution works:
When continuity analysis finds critical or major issues (like a character named "Emma" in Book 1 but "Emily" in Book 3), the resolution wizard lets you fix them before editing starts.
- Smart defaults — the most frequent variant is suggested as the canonical name
- Per-issue control — approve, skip, or change the resolution for each issue
- Manuscript-level fixes — approved changes are applied across all affected books and chapters
- Non-destructive — original files are preserved; resolved copies are used for editing
Writing with Assist
The editing pipeline is for manuscripts you've finished. Assist is for the chapters you haven't written yet. It reads what you have, helps you plan where the story goes next, and drafts new chapters in your voice — but nothing lands in your manuscript until you review and accept it.
When Assist Is For You
- You're partway through a book and stuck on what happens next
- You have a chapter's shape in your head but want a full first draft to react to
- You want to keep writing on schedule without sacrificing your voice
- You're drafting a sequel and need the AI to stay consistent with the prior book
If your manuscript is done and you want it edited, use the four-stage pipeline instead — see The Four Editing Stages.
Where to Find Assist
Open any project from your dashboard. From the project page, click Assist. That takes you to the Assist dashboard — the home base for all writing tools. Every Assist page has a ? button in the top-right that opens a contextual help panel for that specific screen.
Two Setup Steps (Do These First)
Extract your style fingerprint
Before the AI can write in your voice, it needs to learn it. On the Assist dashboard, click Extract My Style. The system reads the chapters you've uploaded and measures 19 specific things about your writing — sentence rhythm, paragraph length, vocabulary habits, dialogue patterns, punctuation choices, and more. Takes about 60 seconds.
The result is a style fingerprint — a quantitative profile of how you write. You only need to do this once per project. If you later add a lot of new writing, re-extract to keep the fingerprint current.
Run the Story Brain analysis
The Story Brain is a map of your book's world. From the Assist dashboard, click Analyze Manuscript. Claude reads every chapter and extracts your active characters, open plot threads, story beats, world rules, and the gaps it notices — things that got set up but never paid off, or facts that stopped being mentioned. Takes 1-3 minutes depending on manuscript length.
The Story Brain is what lets the planner give you chapter options that actually fit your story instead of generic suggestions. Re-analyze anytime you've accepted significant new content — the Brain should reflect the current state of your manuscript.
The Writing Loop
Once setup is done, you can write new chapters using this four-step loop. Repeat it as many times as you need.
1. Plan — where does the story go next?
On the Forward Planner, pick the chapter you want to work on, then click Generate 3 Options. Claude reads your Story Brain and your most recent chapters, and proposes three distinct narrative directions — each with its own pacing, focus, and consequences.
Optionally, fill in the Your direction field first. This isn't a filter — it's a steering input. All three options will be built around whatever you write. Be specific. "Something romantic" produces generic options; "Caleb finally learns what Maya is hiding, but it costs him" gives the AI a real dramatic problem to solve.
Pick one option. The planner breaks it down into a beat plan — numbered scene-level beats that outline how the chapter will unfold. You can either draft the full chapter from here, or draft beat-by-beat if you want tighter control.
2. Draft — the AI writes in your voice
Click Draft Full Chapter. The style enforcer writes the chapter, scores it against your style fingerprint, and rewrites it up to three times trying to hit an 80% voice match. You see live progress — "Iteration 2 — 74% style match" — in the tile where you clicked.
Typical time: 1-3 minutes. You can leave the page and come back. When it finishes, the tile swaps to a green "Draft ready — View in Staging" link, and the planner's top strip shows a persistent "Drafts ready for review" banner until you clear the queue.
Prefer scene-by-scene? From the chapter view, you can draft one scene at a time using commands like Continue, Expand beat, Rewrite in voice, or Alternatives. Useful when you know the shape of the chapter and want control at the paragraph level.
3. Review in Staging
Every AI draft lands in Staging — a holding area between "the AI wrote this" and "this is in my manuscript." Nothing affects your book until you click Accept.
Staging groups drafts by chapter, with filter pills at the top (Pending, Accepted, Discarded, All). Each draft card shows a style match percentage and one of two labels:
- Voice target met — the draft hit 80% voice match and the enforcer stopped early. Should blend into your manuscript cleanly.
- Best of 3 attempts — the enforcer ran all three passes without hitting 80%. You're seeing the highest-scoring attempt. Plan to edit toward your voice before accepting.
Click Review & edit to open the draft in a full editor. Make changes directly, click Re-score my edits to see how much you moved the style match, then Accept or Discard.
4. Accept into the manuscript
Accept inserts the draft at the right position in your book and records a provenance entry — a note that this section was AI-generated. Provenance is how your style fingerprint stays honest: when you re-extract your fingerprint, AI-generated passages are excluded so the fingerprint reflects your writing, not the AI's.
Once accepted, the draft is removed from Staging and appears in your manuscript view. From there you can export the full book, run the editing pipeline over it, or go back to the planner and write the next chapter.
Keeping Staging Clean
During an iteration session you'll often generate multiple drafts for the same chapter before picking one. Staging has three affordances for keeping the list manageable:
- Filter pills at the top — default Pending hides accepted and discarded drafts so you see only what still needs a decision
- Per-chapter "Discard N pending" — when a chapter has 2+ pending drafts, a header button lets you clear the iteration noise after you've picked the one you like
- Permanently delete discarded — on the Discarded filter, a footer button removes discarded drafts from the database entirely. Accepted drafts keep their provenance, so this only affects the ones you rejected.
Tips for better results
- Re-extract your style fingerprint after you've accepted 3-5 new AI-assisted chapters. Your voice fingerprint stays accurate because accepted AI passages are excluded from extraction automatically.
- Re-run the Story Brain after major developments (new character, plot reveal, chapter acceptance). The planner's suggestions are only as good as the Brain it reads.
- Be specific in the direction field. "Chapter should introduce tension between Caleb and Maya over the missing journal" outperforms "build tension."
- If a draft comes back as "Best of 3 attempts" with a low style match (under 70%), the fingerprint might be out of date or the chapter's subject matter might differ dramatically from your training corpus. Re-extract the fingerprint and try once more.
- Don't feel obligated to accept anything. Drafts are a cheap way to see what a chapter could look like. Discard freely.
The Four Editing Stages
Each stage focuses on a different layer of your writing. Run all four for a comprehensive edit, or pick just the ones you need.
1. Pre-Edit Assessment
Sonnet 4.5 · Report onlyA read-through that produces a detailed quality report without changing your manuscript. Think of it as your editor's first-read notes.
What it does:
- Scores each chapter on pacing, dialogue quality, tension, prose style, and character development
- Identifies structural issues—sagging middles, rushed endings, unearned climaxes
- Flags areas that need the most attention in subsequent stages
Output: Detailed assessment report with per-chapter scores and recommendations. No rewrites.
2. Developmental Edit
Opus 4.6 · Full rewriteThe big-picture edit. This is the stage that actually rewrites your manuscript for structure, pacing, and character arcs — expanded scenes, tightened transitions, strengthened motivations.
What it does:
- Rewrites for structure, pacing, and narrative arc
- Deepens character development and strengthens motivations
- Expands underdeveloped scenes, cuts redundant ones
- Fixes plot holes and improves cause-and-effect chains
- Processes chapters in batches to maintain continuity across scenes
Output: Rewritten manuscript with structural improvements — tightened pacing, clearer motivations, filled gaps.
3. Line Edit
Opus 4.6 · Sentence-level polishSentence-by-sentence refinement. Targets the rhythms and patterns that make AI-assisted writing feel generic, while preserving your voice.
What it does:
- Sharpens voice and tightens prose rhythm
- Fixes dialogue that sounds stiff or expository
- Eliminates AI-tells: overused transitions, generic metaphors, purple prose, "telling" instead of "showing"
- Applies custom anti-AI rules to catch patterns that make assisted writing obvious
- Preserves your authorial voice while removing rough edges
Output: Polished manuscript with a consistent, natural voice — fewer AI tells, tighter prose, cleaner dialogue.
4. Copy Edit
Sonnet 4.5 · Light-touch cleanupThe final polish. Catches the small stuff without changing your voice or style.
What it does:
- Fixes grammar, spelling, and punctuation errors
- Ensures consistency of character names, places, and details across chapters
- Normalizes formatting (em dashes, quotation marks, spacing)
- Catches homophone errors and commonly confused words
Output: Clean, publication-ready manuscript.
Story Structure & Beat Detection
Understand how your story maps to proven narrative frameworks, and optionally use that analysis to guide your developmental edit.
What it is
Beat detection maps your chapters to established story structure frameworks—Save the Cat, the Hero's Journey, Three-Act Structure, and others. Each major story "beat" (inciting incident, midpoint reversal, dark moment, climax, etc.) is located in your manuscript and scored for how effectively it lands.
How it works
- The AI reads your entire manuscript and identifies which framework fits best (or you can choose one manually on the configure page)
- Each chapter is tagged with its structural role—setup, rising action, climax, resolution, etc.
- Beats are scored on presence, timing, and impact
- Weak or missing beats are flagged with specific improvement suggestions
The structure report
You get a visual beat map showing where each beat falls in your manuscript, per-beat scores, and a breakdown of structural weaknesses. This report is available on the results page alongside your edited manuscript.
Beat-guided editing
When beat detection is enabled alongside the developmental edit, the AI actively strengthens weak beats during the rewrite. A flat midpoint reversal gets more tension. A missing "dark moment" gets one built in. The structure analysis feeds directly into the editing process.
Tip: Even if you don't run a developmental edit, beat detection on its own produces a useful diagnostic. Toggle it on during configuration to get the structure report without any rewrites.
Settings & API Keys
Cherry Edits uses a Bring Your Own Key (BYOK) model. Your API key, your provider account, your data. No markup.
Anthropic (Direct)
Connect directly to Anthropic's API. Cherry Edits picks the right model for each operation automatically — you don't configure this per-stage. Key prefix: sk-ant-.
Editing pipeline routing
| Pre-Edit Assessment | Sonnet 4.5 |
| Developmental Edit | Opus 4.6 |
| Line Edit | Opus 4.6 |
| Copy Edit | Sonnet 4.5 |
| Beat detection | Sonnet 4.5 |
| Series bible extraction | Haiku 4.5 |
| Continuity cross-reference | Opus 4.6 |
| Continuity resolution | Sonnet 4.5 |
Assist routing
| Style fingerprint extraction | Haiku 4.5 |
| Story Brain analysis | Haiku 4.5 + Sonnet 4.6 |
| Forward Planner (3 options + beat plan) | Sonnet 4.6 |
| Chapter draft / scene draft | Opus 4.6 |
Note: the editing pipeline uses Sonnet 4.5 for analytical stages, while Assist's Forward Planner and Brain analysis use Sonnet 4.6. This is deliberate — Assist's planning needs the newer model's improved narrative reasoning. You don't configure this; we pick it automatically.
- Estimated cost: ~$13.60 for a full 80,000-word editing run (all 4 pipeline stages) — based on 10 completed edits
- Assist costs vary by chapter length and how many drafts you discard before accepting one — typical per-chapter cost is $0.40–$1.50 on a novel of average length
- Prompt caching can reduce actual costs significantly on re-runs of the same manuscript
OpenRouter
Access 10+ models through a single API. Choose separate models for creative and analytical tasks. Cost varies by model selection.
- Key prefix:
sk-or- - Separate model selection for creative (dev/line) and analytical (pre-edit/copy) stages
- Mix and match providers—use Claude for dev edit, GPT for copy edit, etc.
Managed Mode
Coming soonA managed option where we handle the AI provider billing is in development. For now, all users bring their own API key (BYOK) via the Craft plan.
| Manuscript Size | Time (all 4 stages) | Est. Cost (Anthropic) |
|---|---|---|
| Short novel (30,000 words) | ~57 min | ~$5 |
| Standard novel (50,000 words) | ~94 min | ~$8 |
| Full novel (80,000 words) | ~151 min | ~$14 |
| Epic (120,000 words) | ~226 min | ~$20 |
Costs based on 10 completed edits (472,209 words processed). OpenRouter pricing varies by model. Prompt caching can reduce actual costs on subsequent runs.
Cost per stage (per 1,000 words)
| Stage | Cost / 1K words | 80K novel |
|---|---|---|
| Pre-Edit Assessment | $0.020 | ~$1.60 |
| Developmental Edit | $0.080 | ~$6.40 |
| Line Edit | $0.050 | ~$4.00 |
| Copy Edit | $0.020 | ~$1.60 |
Troubleshooting
My pipeline failed mid-run. Did I lose progress?
The wrong number of chapters was detected.
# or ## headings. You can re-upload the file after adjusting formatting.
My API key isn't working.
- Check the key prefix: Anthropic keys start with
sk-ant-, OpenRouter keys start withsk-or- - Make sure you selected the correct provider in Settings before pasting the key
- Verify your API account has available credits or a payment method on file
- If you just created the key, wait a moment—some providers have a brief activation delay
The pipeline seems stuck.
The cost was higher/lower than the estimate.
Can I re-run a project after downloading results?
Can I run just one editing stage?
What's a "story bible"?
What file formats are supported?
Glossary
| BYOK | Bring Your Own Key. You provide your own API key from Anthropic or OpenRouter. Your manuscript goes directly to your provider account—Cherry Edits doesn't proxy or mark up API costs. |
| Continuity Analysis | A series-only feature that reads all books and cross-references characters, timelines, world-building, and plot threads to find inconsistencies. |
| Continuity Resolution | An optional step after continuity analysis. Fixes naming inconsistencies and other cross-book issues by applying find-and-replace or AI-powered rewrites to the manuscripts before editing begins. Original files are preserved. |
| Story Bible | A comprehensive reference document generated from continuity analysis. Contains character profiles, timeline, world rules, and plot threads for the entire series. |
| Beat Detection | Maps your chapters to story structure beats (inciting incident, midpoint, climax, etc.) from established frameworks like Save the Cat or the Hero's Journey. |
| Beat-Guided Editing | When enabled, the developmental edit uses beat detection results to actively strengthen weak structural beats during the rewrite. |
| Pre-Edit Assessment | A diagnostic-only stage that scores chapters and identifies issues without making changes. Produces a detailed quality report. |
| Developmental Edit | Big-picture editing: rewrites for structure, pacing, character arcs, and plot consistency. The most transformative stage. |
| Line Edit | Sentence-level polish: voice, rhythm, dialogue, anti-AI pattern removal. Makes prose read naturally. |
| Copy Edit | Final cleanup: grammar, spelling, punctuation, and consistency across the manuscript. |
| Batch Size | The number of chapters processed together in a single AI call. Larger batches give better cross-chapter context but use more tokens. The default works well for most manuscripts. |
| Pipeline | The full sequence of editing stages that runs on your manuscript. Progress is saved per chapter, so a pipeline can be paused and resumed. |
| Prompt Caching | An AI provider feature that reduces costs when the same context is sent in multiple calls. Re-running a pipeline is often cheaper than the first run because of caching. |
| OpenRouter | An API gateway that provides access to multiple AI models from different providers through a single API key. |
| Creative Model | The AI model used for operations that rewrite or generate prose (developmental edit, line edit, Assist chapter/scene drafts). Defaults to Opus 4.6 for Anthropic users. |
| Analytical Model | The AI model used for reasoning and assessment (pre-edit, copy edit, beat detection, continuity resolution). Editing pipeline defaults to Sonnet 4.5; Assist's Forward Planner and Brain analysis default to Sonnet 4.6 (newer, better narrative reasoning). |
| Extraction Model | The fast, low-cost model used for lightweight extraction tasks (style-fingerprint measurements, Story Brain first-pass extraction, series bible building). Defaults to Haiku 4.5 — about 15× cheaper than Opus per token. |
| Assist | The writing-assistance side of Cherry Edits (as opposed to the editing pipeline). Helps you plan and draft new chapters in your voice. Found under the Assist button in any project. |
| Style Fingerprint | A quantitative profile of your writing style measured across 19 dimensions (sentence rhythm, vocabulary habits, punctuation, dialogue patterns, and more). Used to score AI drafts so they sound like you. Extract it once per project from the Assist dashboard. |
| Story Brain | A map of your manuscript's active characters, open plot threads, story beats, world rules, and narrative gaps. Built by an AI pass over your full manuscript. Feeds the Forward Planner so suggestions fit your story instead of being generic. Re-run after accepting significant new content. |
| Forward Planner | The Assist page where you plan a new chapter. Generates three narrative options based on your Story Brain, breaks the one you pick into scene-level beats, and hands off to the drafter. Lives at /assist/plan in any project. |
| Beat Plan | A numbered sequence of scene-level beats for the chapter you're about to draft. Produced by the Forward Planner after you pick a narrative option. You can draft the whole chapter from the beat plan, or draft one beat at a time. |
| Style Enforcer | The loop that produces each AI draft. It writes the chapter, scores it against your style fingerprint, and rewrites up to two more times trying to reach an 80% voice match. You see which iteration produced the draft you're looking at. |
| Voice Target (80%) | The style-match threshold the enforcer aims for. A draft labelled Voice target met hit 80%+ and stopped early. A draft labelled Best of 3 attempts never hit 80% within three passes — it's the highest-scoring attempt and likely needs light manual editing. |
| Staging | A holding area for AI drafts you haven't decided on. Nothing in Staging is in your manuscript yet. Drafts are grouped by chapter and filtered by status (Pending, Accepted, Discarded). Review, edit, accept, or discard here. |
| Provenance | A record of which passages in your manuscript were AI-generated. Recorded automatically when you accept a draft. Used to exclude AI-written text from future style-fingerprint extractions so your fingerprint keeps reflecting your writing. |
| Narrative Spec | The detected POV mode, tense, POV character(s), and alternation pattern for your project. Shown on the Assist dashboard under Narrative Voice. Used as a hard constraint when Assist drafts chapters — drafts that drift from the spec are rewritten. You can override the detected values manually from the dashboard. |
| POV Mode | The narrator's vantage point: first person (I/me/my), third-person close (pronouns, one character's interior), third-person omniscient (multiple interiors), or second person (you). Controls which pronouns and interior access the AI uses when drafting. |
| POV Character | The character whose head you're in for a given chapter or scene. AI drafts can only access this character's thoughts and feelings. Other characters are only observable through what the POV character sees, hears, or infers. |
| POV Pattern | How POV rotates across chapters: single (one POV throughout), alternating duet (two characters alternate), rotating ensemble (three or more cycle), dual scene-level (two characters switch at scene breaks within chapters), or free (author picks per chapter). Tells the planner which POV applies to the next chapter. |
| Head-hopping | When the narration accesses a non-POV character's interior state as fact (“Tom knew she was lying” in a Sarah-POV chapter). The validator flags these with line-level evidence and asks the drafter to rewrite. Shown as a red POV: fatal badge on staged drafts. |
Questions? Check Troubleshooting above or reach out on our community page.