Skip to Content
CMPCloud Docs

Knowledge libraries

Convert Markdown collections into structured, source-backed CMP knowledge.

Knowledge libraries

A knowledge library is an organized body of knowledge derived from Markdown files, not simply a collection of files hosted online.

Documents contribute headings, metadata, source line references, related documents, and proposed evidence records. An agent or repository can refer to the library without copying all of its content.

Import or preview a library

Visit Knowledge libraries. Signed-in users can select Markdown files or a folder and preview the evidence structure. The preview does not save or publish your source files.

You can also perform conversion entirely on your own machine:

pnpm import:library ./my-markdown-vault --id research-notes --title "Research Notes"

To explicitly produce the complete draft (including source text) in a local file:

pnpm import:library ./my-markdown-vault --id research-notes --title "Research Notes" --output ./research-draft.json

The output contains sensitive knowledge text. Keep it private, and delete it when you no longer need it. The default command prints only a summary.

What conversion produces

  • Sources: Each Markdown file receives a stable source ID, title, tags, content revision, and file path.
  • Sections: Headings become grouped evidence entries. A section carries a source reference such as markdown:notes%2Findex.md#L5-L12.
  • Relationships: Markdown links and Obsidian-style [[links]] become source relationships, including unresolved links.
  • Review state: Extracted content starts as proposed evidence. The importer does not automatically pronounce any sentence a verified fact.
  • Library revision: A deterministic digest tracks the combined versions of included files.

A library can later be linked to repositories, skills, agent definitions, or other knowledge. A link never grants its target's private permissions.

Security and lifecycle

The current preview handles input in memory and requires verified authentication. Permanent encrypted uploads are not enabled. The planned schema stores encrypted payload references, source provenance, version and review state while keeping plaintext Markdown out of library metadata.

CMP's self-managed key service is intended to support separate encryption domains for libraries and session archives. Cryptographic erasure requires verified key sanitization and backup controls. It is not accomplished by hiding a document or deleting a PostgreSQL row.

On this page