The problem

One Substack essay had to be reshaped for LinkedIn, X, Instagram, Threads, Bluesky and video, then scheduled and published through separate tools. The repeated work also risked sanding the writing down into a generic AI voice.

What it does

A self-hosted content engine that uses a voice profile from your writing to create platform-specific drafts. It publishes directly to Bluesky, X, Threads and LinkedIn; Instagram is prepared for manual posting.

Generation runs through Claude in headless mode by default, so there is no per-call bill and no third-party tool in the loop. It has its own six-page marketing site, live at repurposestudio.vercel.app: self-hosting is free and available today, the hosted tier is planned and labelled that way, and interest in it goes straight onto a waitlist.

Who it is for

Writers, founders and creators adapting long-form writing for social posts.

What it does, feature by feature

Pulls your writing in. Imports essays from Substack automatically, takes anything you paste, or watches your feed and drafts new posts every 30 minutes.
Writes in your voice. Builds a voice profile from your archive and past posts, and keeps learning from the edits you make.
Platform-specific drafts. A LinkedIn post, an X thread, an Instagram caption, Threads, Bluesky and a short-video script, each following that platform's rules.
Sharper drafts. Five alternative hooks per draft, a grammar check that respects your voice, and rewrite-with-feedback.
Visuals. 1080×1350 carousel slides and quote cards, rendered offline.
Narration and video. Reels read in your own cloned voice, generated on your own PC, or a free Microsoft voice. Finished MP4s in 9:16, 1:1 or 16:9 with captions timed to the audio.
Ideas. An ideas inbox, ten essay ideas suggested from your archive, outlines, and remixing two essays into a new one.
Scheduling and publishing. Post now or schedule. A planner spreads approved drafts across the week and posts straight to Bluesky, X, Threads and LinkedIn.
Safety net. Checks every draft before you approve it, retries failed posts, resumes half-posted threads and sends a Telegram alert if something still fails.
Any AI model. The default route runs on an existing Claude plan with no extra per-call cost; your own Anthropic, OpenAI, Gemini or OpenRouter key also works, billed by that provider.
On your phone. Installs as an app, with an Android wrapper, so the desk is reachable away from the PC.

Why it is built this way

I built the whole loop locally: ingest once, generate per-platform drafts, learn from my edits, render the assets and publish through each platform's own API. Generation defaults to the Claude subscription I already have, so there is no per-call bill or third-party scheduler; anyone else running it can bring their own Anthropic, OpenAI, Gemini or OpenRouter key instead, encrypted at rest the same way as the publishing credentials, and the settings screen states plainly who pays. Instagram stays manual because Meta requires a publicly hosted image URL that a private local app cannot provide.

What was hard

Windows path escaping broke ffmpeg's subtitle filter, so video rendering switches to relative filenames and a controlled working directory. The ship-ready pass also had to guard malformed model output after it produced `undefined` drafts and fix missing follower snapshots that rendered as `undefined` in the planner.