File layout on disk

Where everything is written, in a folder you can open, grep and version.

rememd stores everything as plain files in ~/Recall/.

~/Recall/
  .recall/
    config.json            ← global config (active workspace, feature)
    mcp.log                ← MCP server log
  {workspace-slug}/
    index.json             ← all structured data for this workspace
    context/
      {product-id}.md      ← product vision
      {product-id}-knowledge.md  ← accumulated learnings
      {feature-id}.md      ← feature PRD
      workspace.md         ← workspace-level context
    sessions/
      {session-id}.md      ← session log markdown

index.json

Contains:

  • workspaceId, lastModified
  • products[] - name, id
  • focusAreas[] - name, productId, repoPaths
  • features[] - name, status, focusAreaId, repoPath
  • sessions[] - title, featureId, branch, events[]
  • notes[] - content, featureId, productId
  • backlog[] - title, priority, focusAreaId, promotedToFeatureId
  • tasks[] - title, status, featureId, blockedBy[]
  • issues[] - title, severity, featureId, sessionId

All types have backward-compatible decoders using decodeIfPresent so older indexes load cleanly.

Why plain files?

  • No lock-in - ~/Recall/ is just a folder you can inspect, back up, or version-control
  • Fast - no database server, no network, no ORM
  • the agent-readable - if MCP breaks, you can cat the files
  • Git-friendly - could be versioned in the future