Write new posts from your own archive, in your voice, on your machine.
We turn the content you have already published (articles, newsletters, even your videos) into a private writing engine. It is designed to write only from what you wrote, in your voice. Your archive is built and stored locally on your own computer. When you generate a draft, only the prompt and the retrieved excerpts go to your chosen AI provider, under your own account.
See how it works- Archive stored locally
- Offline keyword search
- Works in Claude or OpenAI
- Your archive never enters a git repo
CLI screen-capture placeholder. A short loop of one command and a blog draft appearing (real tool, to be added).
You have already written a lot. None of it is working hard enough.
Your back-catalogue is gathering dust.
Years of articles, newsletters and talks, full of your best thinking, sitting in an archive nobody re-reads. Every new post starts from a blank page, as if none of it existed.
Generic AI sounds nothing like you.
The cloud writing tools all draw from the same anonymous internet soup. The output is fluent, forgettable, and unmistakably not your voice. Your readers can tell. So can you.
You do not want your whole archive sitting in a SaaS knowledge base.
Uploading your entire back-catalogue into a third-party knowledge base may not be acceptable to you. This keeps your archive on your own machine and sends only the selected excerpts for a single draft to your own AI account at generation time.
One private engine, built from everything you have already published.
We do the hard part once: we pull in your existing content, clean it, and organize it into a private knowledge base, built around your topics and your audience. Then you get a simple kit that runs on your own machine. Ask it for a blog post and it writes a real first draft, grounded in your material, in your voice, for one clearly-chosen audience at a time. Every draft lists the sources it drew on, and you review before publishing. Your archive is built and stored locally; only the excerpts for one draft go to your own AI account when you generate. And the engine you own can do far more than blog posts over time.
It is designed to write only from what you already published, in your voice, and to show you which sources each draft drew on.
How it works
We ingest your content (the hard part, done for you).
You point us at where your content lives: your website, your newsletter, your YouTube channel. We pull it all in, clean it up, break it into searchable pieces, and tag each one by topic and audience. You see a preview and approve what goes in, item by item. This is where most DIY attempts fail: extraction, cleanup, tagging, and QA. We do it once, properly.
You get your private kit.
We hand you a small, self-contained kit: your knowledge base plus a few simple commands. It installs on your own computer. No accounts to create with us, no server to rent, no whole archive uploaded to a third-party service.
You write in your voice.
Open your AI tool, type one command like /blogpost, and get a real draft, grounded in your own material, written in your voice, for one audience you choose. The draft cites the sources it used. Edit, review, publish, done.
$ /blogpostTerminal screenshot placeholder. The /blogpost command producing a draft (real screenshot, to be added).
Optional: a research agent grows your corpus.
Add an optional research agent that goes looking for more of your own published material to fold in, and scans for topic gaps worth writing about next. It always asks before adding anything new.
We do the heavy lifting. You just write.
- Pull your content from website, newsletter, YouTube
- Clean, de-duplicate, and structure everything
- Tag every piece by topic and audience
- Build and hand over your private kit (yours forever)
- Optional: set up the research agent
- Optional: periodic re-ingest (Care, never required)
- Type one simple command when you want a post
- Choose the audience for each piece
- Review and edit the draft before publishing
- Publish wherever you like
- Optional: approve new material the research agent finds
- Re-ingest yourself anytime with the included script
No coding. No whole archive uploaded to a SaaS tool. No server to manage. If you can type a sentence, you can run it. Extraction, cleanup, tagging, and QA are where most DIY attempts fall over; that part is on us.
Works in Claude or OpenAI. Portable, no vendor lock-in.
Your kit is built on a portable architecture: one model-neutral brain (your corpus, the tools, the instructions, the prompts) with thin adapters for Claude Code and OpenAI's Codex. If one vendor changes its pricing or terms, you point your kit at the other; your knowledge base, your voice, and your commands stay the same. We validate each adapter against the same queries and prompts before handover, so you keep the same source-grounded, one-audience behavior whichever provider you run. You own the engine; the AI provider is the interchangeable motor.
Your archive stays on your machine. Only the excerpts for one draft are sent at generation time.
Your content archive is built and stored locally on your machine. Keyword search runs fully offline with no keys. When you generate a draft, only the specific prompt and the retrieved excerpts are sent to your chosen AI provider (Claude or OpenAI) under your own account. We never upload your whole archive to a third-party service, and your archive never enters a git repository. For people who do not want their entire back-catalogue sitting in someone else's knowledge base, that is the difference between using AI and staying on the sidelines.
What leaves when you draft
The prompt you typed, plus the handful of excerpts the engine retrieved from your corpus for that single draft. That goes to your own Claude or OpenAI account.
What stays
Your full archive, the index, and your commands, all on your machine. Keyword search never leaves your computer.
of German firms name data protection as their number-one barrier to using AI.
(Bitkom)
cite legal uncertainty as a barrier.
(Bitkom)
cite a lack of internal know-how.
(Bitkom)
prefer German or European AI providers.
(Bitkom)
Needs source (unverified on the cited page): a claim that around 70% have halted innovation projects over data-protection concerns, up from 61%. Not published as a figure here until confirmed against the official Bitkom page.
A truly-offline option (drafting with a local model, for example via Ollama) is on the roadmap for clients who cannot send any text to a cloud provider.
The EU AI Act's transparency obligations take effect 2 August 2026. Keeping your archive local reduces your data exposure, though drafting still uses Anthropic or OpenAI under your own account. (We are builders, not lawyers. While we build your corpus we act as your Auftragsverarbeiter and sign a proper AVV (Art. 28 DSGVO); that is a bounded, defined liability window, not an open-ended one. Legal advice stays with your advisor.)
One corpus. Many outputs.
Your engine is built once and reused. The same private corpus drives new formats over time, against the same material, with no re-ingest.
- Blog posts
- Newsletter
- Social
- Lead magnet
- Course outline
Output packs run against the same corpus, no re-ingest.
Start with one capped, own-forever build.
The Starter Archive Engine is a hard-capped, one-time build. You own it forever and run it on your own machine. Everything beyond it is an optional upgrade, never a subscription you have to keep.
Starter Archive Engine
Your private writing engine from one source of your content. One source type (a WordPress export, an RSS feed, or a sitemap scrape), up to around 50 sources, offline keyword search, one audience or audience-neutral, a voice starter profile, a ZIP plus a written setup guide, one onboarding call, and one correction round. Sized as a focused starter build. Any extra work is quoted before we do it.
Proposal to validate.
Book a free 20-minute fit callAfter fit, optional upgrades
Archive Check
refundable against a purchase
We look at your archive and sources, confirm fit, and scope the build before you commit. The low-friction way in.
Proposal to validate.
Full Corpus Build
own forever
The real bespoke ingest: 2 to 3 source types, cleaning, dedup, audience tagging, voice config, scrub verification, onboarding.
Proposal to validate.
Messy-Source Add-ons
For trickier inputs: scrape quirks, newsletter-platform API access, YouTube channel transcripts, PDF or OCR, manual audience review by volume.
Proposal to validate.
Output Packs
own forever
Newsletter, social, lead magnet, or course outline, against the same corpus, no re-ingest.
Proposal to validate.
Care (optional)
never required
Periodic re-ingest, kit updates, support, and new output packs. Your base kit works forever without it, and ships with a self-reingest script so you can refresh it yourself.
Proposal to validate; monthly only for high-cadence clients.
All figures are proposals to validate, shown as from / typical ranges.
Book a free 20-minute fit call, or get a free Archive Check.Already working in the real world.
sources ingested
searchable pieces
audience per post
archive stays on her machine
Dr. Dina Beneken, Lerncoach (learning-support educator), lernraeume-beneken.online
We ingested 224 sources of Dr. Beneken's real content (newsletters, blog posts and website pages spanning years) into a private knowledge base of 720 searchable pieces, each tagged by topic and, where applicable, by audience: parents (Eltern) and learning-support professionals (Fachpersonen). She now writes new posts on her own Windows computer, in her own voice, for one audience at a time. The engine is designed to keep the two audiences separate and to write only from her material, and every draft shows the sources it drew on. The whole archive lives on her machine; only the excerpts for a single draft go to her own AI account at generation time. The before-and-after is the point: years of scattered, hard-to-reuse content became a single engine she can write from in minutes, instead of starting every post from a blank page.
[Client quote from Dr. Beneken, to be gathered.]
Client photo slot (added once approved)
Questions, answered.
Is this just ChatGPT with extra steps?
No. ChatGPT writes from the whole internet and runs in someone else's cloud. Your kit is designed to write only from your own published content, in your voice, and your archive stays on your own machine. When you draft, only the prompt and the retrieved excerpts go to your own AI account. It is the opposite trade-off: narrow, private, and unmistakably yours.
Do I need to be technical?
We do all the setup. Day to day, you type one simple command and get a draft. It is a guided command-line tool, so it works best if you are comfortable using one, or you can choose a done-with-you or managed option if you would rather we drive.
What actually happens to my data?
Your content archive is built and stored locally on your machine. Keyword search runs fully offline with no keys. When you generate a draft, only the specific prompt and the retrieved excerpts are sent to your chosen AI provider (Claude or OpenAI) under your own account. We never upload your whole archive to a third-party service, and your archive never enters a git repository. While we build your corpus we act as your Auftragsverarbeiter and sign a proper AVV (Art. 28 DSGVO), and we delete every working copy and credential after delivery.
Claude or OpenAI, which do I need?
Either. The kit is built on a portable architecture and runs in both, so you are not locked to one vendor. We will help you pick based on what you already use, and we validate the behavior on each before handover.
Will it invent things or make claims I did not make?
It is designed to write only from your corpus and to flag when a topic is not covered. Every draft lists the sources it drew on, and you review before publishing. Like any AI tool it is not a zero-error guarantee, which is why review is part of the flow.
I write for two very different audiences. Will it mix them up?
It is designed to write each asset for exactly one audience you choose. You pick the audience per piece, and review the draft before it goes out.
Do I have to pay every month to keep it working?
No. The base kit is yours forever and works without any recurring payment. It ships with a self-reingest script and written instructions, so you can refresh your corpus yourself when you publish new content. Care is an optional convenience for people who would rather we handle the refresh.
Let's see what your archive can do.
Tell us where your content lives and what you would love to publish more of. In a free 20-minute call we will look at your material, tell you honestly whether this is a fit, and scope what your Starter Archive Engine would involve. Prefer to start even smaller? Ask for a free Archive Check first.
Prefer email? Write to us at [contact email], we usually reply within a day.
Archive stored locally by default. Your archive never enters a git repo. Only the excerpts for one draft are sent to your own AI account at generation time. We are builders, not lawyers, legal advice stays with your advisor.