I
Where books begin
Three doors into the studio. Write one sentence yourself. Let the trend scout bring you a subject the world is already moving toward. Or let another machine commission the whole thing while you sleep.
Local AI book studio
Tomevell plans, researches, writes, fact-checks, edits and binds complete books — fiction and nonfiction — using local AI models. Nothing leaves your desk unless you say so.
Three doors into the studio. Write one sentence yourself. Let the trend scout bring you a subject the world is already moving toward. Or let another machine commission the whole thing while you sleep.
The one rule everything follows
Tomevell decides what work must be done.
The model router decides which model does each piece.
A book is not one long prompt. It is a hundred small, checkable jobs: a thesis, an outline, research questions, evidence sorted to sections, drafts, transitions, critiques, rewrites, a binding pass. Tomevell runs them as a durable pipeline — every step recorded, every step resumable.
Thirty seconds, the whole journey:
Long books fail in quiet ways: characters knowing things they haven't learned yet, state that resets between chapters, claims that sound right but aren't sourced. Tomevell treats all of them as engineering problems, not hopes — and its repair engine can take an already-finished book back through the same discipline, chapter by chapter.
Who is alive, what broke, what cannot un-happen — folded forward into every next scene. Contradictions are flagged by code, not hoped away.
Every finished book is read whole by independent critics and scored on the shelf. Below the bar, the weakest chapters go back to the desk.
Every draft, rewrite and edit is a kept revision. The whole book has a history you can walk.
The acceptance test kills the process mid-sentence. The book resumes exactly where it died — and never pays twice.
Which model wrote what, every token, every second, every catch — the full provenance, kept.
No model is hired on reputation. Before a book is written, candidates audition blind — same brief, hidden names — and are scored against each other. The best writer for this book gets the chair. Sometimes the small model wins; in our last audition, it did.
Tomevell runs on your hardware against local models. A book can be produced with the network switched off — and a local-only setting makes that a promise, not a preference: the manuscript never leaves the machine. Cloud models are an option you grant per book, never a dependency.
The end of the pipeline is not a text file. It is a book: front matter, chapters, sources and notes, bound to EPUB, DOCX and PDF with checksums and the exact toolchain recorded — ready for a store, an editor, or your shelf.
Everything a person can do in the studio, an agent can do through the API — REST, MCP, or the command line. Your automation commissions a book, polls its progress, approves its gates, and collects the files. It never needs to know which models did the work. That decision belongs to the house.
// One call. The studio takes it from here. POST /v1/books { "type": "nonfiction", "prompt": "How autonomous agents change online publishing", "target_words": 55000, "quality": "balanced", "autonomy": "autonomous", "budget_usd": 20 } // → { "project_id": "TV-2026-0041", "status": "planning", ... }
Not every headline deserves a book. The scout watches world signals and climbs a strict ladder — news, event, trend, structural change — keeping only subjects with years of shelf life. What survives arrives as a worked-out proposal: subject, reader, angles, and a recommendation.
Tomevell is in private development. If you want early access — as an author, a publisher, or someone whose agents should be writing books — say hello.
contact@tomevell.com