AI is powerful. It just doesn't know you. Every session, you start from zero. Upload docs, PDFs, DOCX, code, CSVs & Relatch converts them into a Skill File: amd that teaches AI exactly how you think, write & build. Claude: 5 templates : Persona, Codebase, Process, Domain & Finance. Codex: 3 shapes : Execute, Expertise & Specialist. Every output quality-scored. No generic text. Download amd for Claude or a ZIP for Codex CLI. Your AI remembers you now.
Relatch didn't start as a skill file generator. It started as a complete failure.
V1.1 was a content re-engagement tool , the idea was simple: people save articles & insights but never revisit them. I'd build the nudge layer that brought them back. I shipped it. Users didn't return. Not to the content, and not to the product.
Back to zero.
But sitting with that failure, I realized I'd been solving the wrong layer. The problem wasn't that people forgot what they'd saved. It was that every tool they used especially AI forgot them. Every Claude session started blank. Every Codex run had no idea how they wrote, how they built, what they cared about.
Years of accumulated expertise. Hundreds of documents. Zero memory transfer.
That's what Relatch became: a converter that takes what you've already written style guides, codebase files, process docs, brand decks, financial CSVs and turns it into a Skill File. A structured .md that installs into Claude or Codex and actually teaches it how you work.
The hardest part wasn't the idea. It was the output quality.
Early versions produced exactly what you'd expect from an AI asked to "capture your style": "Focus on users. Think long term. Communicate clearly." Useless. Could apply to any professional on earth.
So I built a 9-point quality scoring system. An anti-abstraction rule. Template-specific required sections. If the output could apply to anyone, it gets rejected and the model tries again. That single decision changed everything about what Relatch produces.
The other thing I didn't anticipate: Claude & Codex need completely different cognitive profiles. Claude learns your persona. Codex needs an execution playbook. So I ended up with 5 Claude templates (Persona, Codebase, Process, Domain, Finance) and 3 Codex shapes (Execute, Expertise, Specialist) each with its own prompt architecture and scoring logic.
It evolved from "help people revisit content" to "encode your expertise into the AI you already use every day."
Would love to hear: what knowledge do you wish your AI already had about you? That's the problem I built this for curious where it lands for you.
Report
No reviews yetBe the first to leave a review for Relatch