The idea
AI assistants are good at organizing what you tell them, but what they learn about you usually stays inside one vendor's product. CyberBible flips that. Plain Markdown files are the single source of truth, stored in a space you control, and you can read and edit every file yourself. Any AI tool that speaks the Model Context Protocol (MCP) can read or write them, only with access you grant and can revoke.
Design principles
- Yours, in your cloud. Each user gets an isolated space in their own cloud account.
- Model-agnostic. One MCP server, so tools such as Claude Code, Cursor or ChatGPT can connect without lock-in.
- Readable by humans. Markdown, not a hidden database. The search index is derived from it and protected the same way.
- Encrypted by default. Per-user keys with envelope encryption; content is decrypted only in memory while being processed, and every decryption is logged.
- History you can undo. Every file keeps a version history, so edits from you and from AI tools can be merged or rolled back.
Planned pieces
| Piece | What it does |
|---|---|
| Onboarding | A guided AI questionnaire (or a call, with consent) drafts your first structured Markdown notes. |
| MCP server | Scoped, revocable tokens for read or write access per tool. |
| CLI | A thin command-line client built on the same MCP interface. |
| Version history | Per-file history, mergeable concurrent edits, rollback. |
| Search index | Vector index encrypted and isolated like the source files. |
Status
This is a concept and technical plan, not a launched product. There are no users and no results to report yet. It grows out of my work at Cible Global, where I built a Python prototype on the Claude API that turns unstructured notes into organized Markdown knowledge.