A small team building one thing: the shortest path from "I own this material" to "my AI knows it cold." No fake team grid. No investor logos. Here's what we believe and why we built this.
Generic AI answers sound authoritative right up until one turns out to be invented. Every quote, every fact, every claim in a skill traces back to your source before it ships. If a line can't be verified, it doesn't go in. We rebuild until every claim is grounded — or we refund your credits and tell you why.
The problem isn't the model — it's that the model doesn't know what you know. Grounding it in your material is what fixes the hallucination problem for the domains that matter to you.
Your source file is used only to build your skill. It is never used to train any model, never retained beyond the build window, never shared or sold. The skill itself lives in your account, encrypted, and only you can access it. Cancel and you keep every skill you built — they don't expire.
We make this promise first and loudest, because the Commoditization-Anxious expert asking "won't uploading my material just leak it into training data?" deserves a clear, categorical answer: no, full stop. We confirm the upload belongs to you before we touch it.
SKILL.md is an open format — adopted by Anthropic, OpenAI and Google. We build your skill once to that standard; it works in ChatGPT, Claude, Gemini and Codex. If you stop using SkillPill tomorrow, you still have the file. You own the artifact. We just built it.
There is no "SkillPill format" that locks you to us. The free, open-source book-to-skill tool can read the same files. We're the convenience layer, not the moat.
2026 was the year skills became cross-platform. Anthropic shipped Claude Skills, OpenAI's Custom GPTs adopted the same SKILL.md frontmatter, Google's Gems followed. A single skill pack now works in all three — plus every agent framework that reads the standard.
Most professionals don't know it yet. And most of the tooling assumes you can read a YAML file and set up API keys in a terminal. That gap — between "the standard exists" and "a non-developer can actually use it" — is the whole company.
The free OSS tool that powers us works perfectly if you're comfortable with Python and token counting. We've paid that forward — it's open source and we'll point you there if it fits. Most people want the skill, not the configuration project. That's what we sell: the outcome, not the toolchain.
STANDARD / SKILL.md2026Anthropic → Claude Skills (native) OpenAI → Custom GPTs (Knowledge) Google → Gemini Gems + 32 agents → AutoGPT, Continue, Cursor… # One pack. Portable everywhere. # The convenience is the product.
A friend asked: "I have 14 books I keep re-summarising into ChatGPT every session. Can we just make ChatGPT know them?"
That question turned out to be the whole product. Three weeks later there was a working CLI tool. Three months after that it was being used by lawyers, founders and ops managers who hated Python. The pattern was the same everywhere: they all wanted convenience, not a configuration project. They wanted to drop a document in and get back a skill that worked — without reading a README.
So we built the hosted version, stripped out every step that wasn't upload → confirm → install, and shipped it. That's SkillPill.
We're a small team — not a funded startup with a marketing budget and a "team" page full of stock photography. The product is the point. We ship things, we fix things when they break, and we're honest about what we don't know yet. The numbers in the sidebar are real; when they change we update them.
The one thing we won't do is invent a grander origin story. The actual origin is: someone had a tedious problem, we built the shortest path to solving it, and it turned out a lot of other people had the same problem.
A lot of people arrive expecting one of the things below. The honest answer is in each card. If any of these are actually what you want, the right tool for that job isn't us — and we'll say so.
SkillPill doesn't replace your AI. It doesn't have a chat window. It doesn't answer questions directly. It builds the briefing pack that your existing AI — ChatGPT, Claude, Gemini — reads before answering. The conversation still happens there. We just make it vastly better-informed.
Fine-tuning changes the model's weights permanently using your data. SkillPill does nothing of the sort. Your source material is never used to train any model. A skill is a structured text file — the AI reads it as context before answering, then forgets it when the session ends. Nothing permanent. Nothing that affects other users.
A prompt pack is generic instructions dressed up for resale: "act like an expert in X, use these frameworks." A skill is built from your actual material — your citations, your methods, your page numbers. It is specific, not template-generic. If the source is yours, the skill is yours. There is no version of this that works the same way for everyone.
Retrieval-Augmented Generation requires chunking, embedding, vector stores, and ongoing eval work. It's the right choice for huge living corpora maintained by an engineering team. A skill is a structured SKILL.md file — engineered for a fixed body of knowledge, readable by any AI that supports the standard, zero infrastructure to run. Different tool for a different job.
