GitWriter
Write once. Version everything. Publish everywhere.
Every book, manual or documentation set becomes a real Git repository — with branches, commits, diffs, blame, merges and conflict resolution — driven entirely from a browser by people who will never install a Git client. It is then published to PDF, EPUB, Word, HTML and a searchable static site from the same source.
13Book types
7Git hosting targets
264REST endpoints
2708Automated tests
The problem
Docs-as-code works beautifully — for the people who can already use Git.
Engineering wants version control, branches and review. The technical writer, the compliance officer and the trainer want to open a page and type. So the documentation splits: the parts under version control that only engineers touch, and the parts in a wiki that nobody can diff, review or roll back. Both halves go stale, in different ways.
01
Two systems
The repository and the wiki disagree about what is current.
02
No review
Wiki pages change with no branch, no approval and no way to see what moved.
03
Publishing by hand
The customer PDF is exported, reformatted and emailed by one person.
04
Content leaves
Hosted authoring puts unreleased specifications and policies outside your control.
The solution
Give everyone the browser. Give the content real Git underneath.
01
Genuine version control
A real repository per book, with branches, annotated tags, merges, three-way conflict resolution, per-file history and blame — driven entirely from the interface.
02
Review before publish
Branch, comment on a line, suggest a replacement, request changes, approve, merge — the workflow engineering trusts, made usable by non-engineers.
03
On your own hardware
Eight of the fifteen supported AI model runtimes are local, and grammar checking runs inside the application — nothing has to leave the network.
Capabilities
01Write & version
- Real, non-bare Git repository per book, scaffolded with a table of contents, sections and an asset folder.
- Commit, revert a commit, paged history, per-file history and blame.
- Branches typed as main, draft, review or archive, with protection and an undeletable default.
- Annotated tags; merges with a dry-run check before you commit to one.
- Book-wide find and replace with case, whole-word and regex options — replace-all lands as a commit.
- Diff between any two references or a single commit, with formatted hunks and rename detection.
- Rich, source and split editor modes with a draggable divider and debounced live preview.
02Collaboration & review
- Live co-editing with per-user cursors and a presence bar showing who is in the book.
- Comments anchored to the quoted text, not just the line number, so highlights survive edits above them.
- Suggested edits — propose a replacement; accepting writes and commits it.
- Review requests from branch to branch with approvals, change requests and threaded comments.
- In-browser conflict resolution — theirs, ours and an editable merged pane per file, with the option to abort.
- Mentions with autocomplete, and in-app notifications for mentions, comments, assignments and jobs.
- Readability scoring, daily word goals, a streak counter and a thirty-day activity sparkline.
03Import & migrate
- Import from Confluence, Notion, GitBook, Markdown, Word or rich text, with the format detected automatically.
- GitBook imports follow the existing table of contents in order, then append anything unreferenced.
- Existing pages are never overwritten — slugs deduplicate instead.
- Unsupported files are skipped with a warning, rather than failing the whole import.
- Bounded and safe: page and file ceilings, archive-entry limits, and traversal defences on every path.
04AI assistance
- Fifteen model providers, eight of them local, so drafting, translation, search and review need no external call.
- An agentic writer plans, drafts, reviews, refines and finalises on its own branch, committing each page as it goes.
- Six path-validated tools for the agent: list, read, edit, create, update content, finish.
- Selection actions streamed token by token: check, rewrite, improve with an instruction, shorten, lengthen.
- Book-level AI work: cross-chapter continuity checking, style linting against your own guide, summaries and glossaries.
- Eight insight tools including pacing, fact-check, link suggestion, alt text, tone and a character bible.
- Semantic search with no vector database — embeddings stored inline, degrading to keyword scoring when unavailable.
05Publishing & platform
- Five output formats from one source: PDF, EPUB, Word, HTML and a static site.
- Export presets inherited across three levels: page size, margins, base size, fonts, contents depth and numbering.
- A cover designer with image, colour, subtitle and author overrides, plus front and back matter.
- The static site includes sidebar contents, a page outline, previous/next navigation, a light/dark toggle and full-text search.
- White-label branding with field-level inheritance — name, logo, favicon, colours and typeface, shown to readers too.
- Reader analytics: views, sessions, searches, helpfulness, per-page figures and drop-off pages.
- Content health checks for broken internal links and orphan pages, recomputed on every build.
06Security & governance
- Two-dimensional access control — a system role and a per-book role — evaluated in one place
- Privilege-escalation guard: nobody can grant a role above their own
- Restricted books that cut off implicit organisation-wide access and vanish from listings
- Tenant isolation with cross-tenant identifiers answering not found rather than forbidden
- bcrypt at strength twelve; brute-force protection with exponential lockout on login, registration and refresh
- Refresh-token rotation with a server-side revocation store
- Single sign-on through Google and Microsoft, configurable at all three levels
- Content security policy, frame denial, strict transport security and referrer policy; specification browser denied by default in production
- Path-traversal defence on every content, tool and branch path; outbound-address guard on administrator-supplied endpoints
- Restore hardened against archive traversal and repository hooks
■Java 17 on Spring Boot 4, with an embedded Git engine — no Git binary or external service.
■PostgreSQL database, 30 tables, schema installed by the setup wizard.
■264 REST endpoints across 39 controllers; server-rendered vanilla ES6 front end, no build step.
■2,708 automated tests — roughly the size of the production codebase itself.
■Single sign-on via Google and Microsoft; bcrypt password hashing; refresh-token rotation.
■Self-hosted, single-node deployment; not yet available: outbound e-mail, MFA, SAML/directory login, encryption at rest for provider credentials.
Industries
Who it works for
Software & product teams
Product docs, API references, specs, runbooks and internal wikis: the docs-as-code workflow, usable by people who are not engineers.
Regulated & air-gapped
The entire stack, including the language model and the grammar checker, runs on-premise with no outbound call.
Compliance & quality
Policies, procedures and specifications where every change needs a branch, an approver and a history you can produce on request.
Documentation agencies
Tenant and organisation separation with per-scope branding and custom domains, so many client brands run from one instance.
Managed service providers
Client runbooks and handover documentation, versioned and published under the client's own domain.
Training & education
Courses and workbooks published as PDF and EPUB, with reader analytics showing which page people abandon.
Manufacturing & engineering
Manuals and service documentation issued as tagged versions, so the field always knows which revision it has.
Publishers & authors
Continuity checking, character bibles, pacing analysis, writing goals and a cover designer for long-form books.
Leaving a hosted wiki
Importers for Confluence, Notion and GitBook archives with automatic format detection, so migration is not a separate project.
Use cases
Example scenarios
01 · Manufacturing & engineering
Service manuals the field can trust to be the current revision
- Challenge
- A machine builder keeps its service manuals as Word files on a shared drive; one person exports the PDF and emails it to technicians. Nobody can say which revision a technician has on site or what changed since the last one.
- Approach
- The Word manuals are imported into GitWriter, each one its own Git repository. Changes are made on a branch and merged after a review request is approved; each release is an annotated tag with release notes, published as PDF and as a searchable site on the company's own domain with an RSS changelog.
- Result
- Every manual in the field maps to a tagged version with a readable diff to the previous one, and publishing no longer depends on one person exporting files by hand.
02 · Compliance & quality
Quality procedures with an approver and a history for every change
- Challenge
- An organisation preparing for an ISO or NIS2 audit keeps its policies and procedures in a hosted wiki. Pages change without approval, and showing an auditor who changed what and when means digging through revision lists.
- Approach
- The Confluence export is imported with automatic format detection, and GitWriter runs on the organisation's own server with a local model and built-in grammar checking. Protected main branches mean every change goes through a review request with approvals or requested changes, and per-file history and blame show who wrote each line.
- Result
- Each procedure change has a branch, a named approver and a permanent history that can be shown to an auditor directly, and the content stays on the organisation's own hardware.
03 · Training & education
Bilingual course material from one source
- Challenge
- A training provider maintains its courses in Dutch and French as separate documents. The two versions drift apart, and nobody knows which chapters learners actually read.
- Approach
- The Dutch course is written in GitWriter and translated as a whole book onto its own branch, page by page; a French-speaking trainer reviews it with anchored comments and suggested edits before merging. Both editions are published as PDF and EPUB, and reader analytics show views, searches and drop-off pages.
- Result
- Both language versions come from one repository with a visible diff between them, and the trainers can see where learners stop reading.
Licence
Per deployment. Never per seat.
Licensed per installation, not per author, so one licence covers unlimited authors. The annual fee scales with your own turnover — marginal like income tax, with a €900 minimum and yearly increases capped at 25% — while hosting, AI usage and services are billed separately.
Annual licence by turnover, ex VAT
| Turnover | Per year |
|---|
| Up to €500k | €900 |
|---|
| €1M | €1,500 |
|---|
| €5M | €2,500 |
|---|
| €25M | €5,500 |
|---|
| €100M | €13,000 |
|---|
| €500M | €37,000 |
|---|
| €1B | €52,000 |
|---|
Illustrative: a seat-priced platform at €12 per author a month bills two hundred writers €28,800 a year; at €5M turnover this is €2,500.
What it does
- 01Real Git — branches, merges, diffs and blame — used entirely from the browser.
- 02Fifteen AI model providers, eight local, so drafting and review never have to leave your network.
- 03One licence, unlimited authors — publish from one source to PDF, EPUB, Word, HTML and a static site.
At a glance
CategoryAuthoring & Publishing
LicencePer deployment, unlimited users, from €900 a year
For whomEngineering docs · Compliance · Training · Agencies · Publishers
Export your wiki.
Give us the Confluence, Notion or GitBook archive. We will import it, publish it, and hand you the site and the PDF.Start the conversation →