Source custody

Your data stays legible

Original evidence, normalized text, generated artifacts, and revisions remain distinct.

Workspace layout

<workspace>/
  sources/<record-id>/
    source.original.<extension>
    source.txt
    source-manifest.json
  artifacts/<record-id>/
    yammerknit.md
    artifact-manifest.json
    validation.json
  revisions/<record-id>/<revision-id>/
    yammerknit.md
    revision-manifest.json
    validation.json

Manifests record relative filenames and hashes, so the complete workspace can move without breaking its internal references.

Default data-handling habits

  • Existing records are not overwritten.
  • Original bytes remain distinct from normalized UTF-8 text.
  • Revisions identify their parent artifact.
  • inspect-record determines the current artifact from immutable manifests.
  • Unknown metadata remains N/A or null.
  • Credentials, browser cookies, tokens, and unrelated page data are not collected.

Network behavior

Core Yammerknit creation uses only the source supplied to the agent. The optional YouTube helper invokes the separately installed yt-dlp program and contacts YouTube for accessible subtitle data.

Yammerknit includes no telemetry service, cloud storage, embedding service, or vector database. The subtitle helper disables browser-cookie loading, playlist traversal, audiovisual download, and local yt-dlp configuration.

You choose the destination

You choose the workspace and any later transfer, embedding, backup, or publication destination. Installing Yammerknit does not upload existing records.

Before sharing support evidence: remove private source content, credentials, and unrelated personal data from logs and receipts.