Named after Daedalus (Δαίδαλος), legendary mortal master craftsman of Greek myth — architect of the Labyrinth, sculptor whose statues "seemed to live and move," builder of the wings for Icarus. The English word daedal means skillfully made. Where the Muses give divine inspiration, Daedalus represents human craft — the patient assembly of intricate, durable things. Building a long-form nonfiction book is daedalan work: many parts, each fitted with care, into a structured whole. Fitting for a tool that helps a human author write their book.
How it works
Daedalus clones the author's voice across 10 dimensions, plans an agency-appropriate book outline, drafts chapter by chapter under a 5-layer context stack that keeps long-range structure coherent, runs a 7-pass revision pipeline (five MiMo passes, a free regex-based AI-slop scan, and a final Anthropic Opus polish), and exports KDP-ready PDF / DOCX / ePub.
A word about how to use this
Daedalus is a drafting and revision assistant, not a ghostwriter. It does the mechanical parts of writing a long book — voice consistency across 80,000 words, continuity across a dozen chapters, multi-pass revision, print-ready formatting — so a human author can spend their time on the things only they can do: the what and the why of the book.
Every chapter it produces is a starting point. You read every sentence. You make it yours. You own every claim in the final manuscript. The slop scanner and the polish pass are mechanical aids — they catch the AI-shaped patterns and rough edges a human editor would also catch, but they don't substitute for your editorial judgment. Treat Daedalus the way you'd treat a very capable ghostwriter whose drafts you still have to rewrite: helpful, fast, and absolutely not the signatory on the title page.
MCP tools
analyze_voice(sample)— extract a 10-dimension voice fingerprint from 3K+ words.plan_book(concept, category, voice, audience, research_context?)— chapter-by-chapter outline with word targets.write_chapter(chapter_number, ...)— draft a single chapter under the 5-layer context stack.revise(target, revision_focus?)— run the 7-pass pipeline on a chapter or the full manuscript.get_manuscript()— retrieve the assembled manuscript and progress.export_book(formats)— export to markdown / PDF (KDP trim sizes) / DOCX / ePub.
Book categories
Pick one — plan_book selects the matching structural template:
| Category | Shape |
|---|---|
| leadership | hook → frameworks → application → edge cases → integration → CTA |
| self_help | wake-up call → problem → root cause → methodology → going deeper → setbacks → transformation → 30-day plan |
| academic_lite | puzzle → conventional wisdom → evidence chapters → synthesis → implications → unknowns |
| business | opportunity → landscape → framework → implementation → case studies → scaling → future |
| memoir_hybrid | pivotal scene → before → journey → crucible → other side → universal truth → forward |
| how_to | why → prerequisites → step-by-step → integration → advanced tips → resources |
The 5-layer context stack
Each write_chapter call assembles all five layers before drafting so chapter N stays coherent with chapters 1..N-1:
- Thesis — the book's core argument. Never changes.
- Sources — research context (e.g. from hyperion or mnemo).
- Outline — windowed: previous chapter, current chapter (full detail), next chapter.
- Voice — the cloned profile or a preset.
- Canon — running continuity: terms defined, examples used, frameworks introduced, unfulfilled promises. Auto-extracted after each chapter, auto-compressed every 5 chapters once it crosses ~6K tokens.
The 7-pass revision pipeline
| # | Pass | Backend | What it does |
|---|---|---|---|
| 1 | structural | MiMo | Argument flow, section transitions, chapter coherence. |
| 2 | content | MiMo | Claims supported, no contradictions, evidence strength. |
| 3 | voice | MiMo | Drift back toward the voice profile. |
| 4 | clarity | MiMo | Readability, jargon reduction, paragraph length. |
| 5 | line | MiMo | Word choice, rhythm, redundancy, weak verbs. |
| 6 | slop | regex (free) | AI-ism scanner · Tier 1 / Filler / Weasel / Structural. |
| 7 | polish | Anthropic Opus | Final quality judgment, remaining issues, score 0.0–1.0. |
Notable design notes
- Voice cloning is per-author, not per-book. One
voice_idcan be passed to manyplan_bookcalls — the same author writes in their own voice across many books without re-analyzing. - Slop runs on every draft automatically. Not just in revision. If a draft crosses threshold, the agent sees a warning before the user does.
- Canon compression keeps long books from blowing the context window. Running canon auto-compresses every 5 chapters once it crosses ~6K tokens, preserving terms / examples / frameworks / promises while merging redundancy.
- Multi-model cost split. MiMo for the high-volume work (drafting, 5 of 7 revision passes), Claude Sonnet for targeted analytical work (voice, canon), Anthropic Opus for the single place quality matters most (the polish pass). Keeps a 50K-word manuscript affordable.
Stack
- FastMCP (Python)
- Anthropic Opus 4.7 (polish)
- Claude Sonnet (voice + canon)
- OpenRouter
- MiMo-V2-Pro (drafting + revision)
- pandoc + tectonic (PDF / DOCX / ePub)
- book-class LaTeX (KDP trim sizes)
- Docker Compose
- LibreChat (host)
Typical chain for a book: hyperion investigates the field → daedalus drafts chapter by chapter → aletheia verifies every claim against mnemo before export.