Vision: The Era of Intent
For decades, open source meant shipping static bytes. You wrote a library, pushed it to a registry, and someone else installed those exact bytes. That worked brilliantly when humans were the compilers.
But that era is ending. AI agents don't need your pre-packaged code; they need your intent. When you tell an agent what to build and how to behave, it writes the code itself — adapting to the local stack in ways a static library never could.
Right now, the best developers are already working this way. But their most powerful instructions are trapped in private notes or buried in chat histories. Open Prompt Hub exists to give those prompts what code got decades ago: a public registry with versioning, discovery, and community trust.
Write an open-prompt
A Markdown file with structured metadata. No code, just instructions: what to build, which models it likes, and how to verify the result.
Publish & Scan
Push it to the registry. Every submission is automatically audited by specialized security models to flag data exfiltration, injection attacks, or destructive patterns before they ever reach another dev's machine.
Fork & Personalize
Prompts are “soft” infrastructure. If a prompt is 90% of what you need, just fork it and append your local rules. Adapting a prompt to your specific project style is as easy as editing a text file.
NPM ships tarballs. We ship intent. Because the AI generates fresh code every time, the output is never “out of date” — it evolves with the models and the environment.
Code is brittle; prompts are fluid. In the Hub, “personalization” isn't a feature request — it's the default. Take a high-performance prompt and instantly tune it for your specific project constraints.
We replace “blind trust” with automated guardrails. By combining AI safety model audits with real-world Build Reports, we provide a heat map of which instructions are safe and effective across different models.
No walled gardens. If an instruction is worth sharing, it should be findable and safe. We believe the best AI instructions should be as accessible — and as secure — as the best open-source libraries.
The registry is step one. We're building toward a future where prompts are first-class packages:
Read the docs to learn the format, then publish your first open-prompt.