Memory Schema

Every piece of information Mindr stores is a MindrMemory. Understanding the schema lets you write precise queries and custom integrations.

MindrMemory

interface MindrMemory {
  id: string                        // stable UUID
  content: string                   // the text of the memory
  role: 'user' | 'system'          // 'user' = manual; 'system' = auto-detected
  sessionId?: string                // agent session that created this memory
  tags: MindrTag[]                  // key/value labels for filtering
  metadata?: Record<string, unknown> // structured data beyond text
  createdAt: string                 // ISO 8601 timestamp
  deletedAt?: string                // set on soft-delete; excluded from queries
}

interface MindrTag {
  key: string
  value: string
}

Memory types

The type tag is the primary classifier. Every memory has exactly one type tag (except metering records which are internal).

Type	Who creates it	Purpose
decision	Post-commit hook, mindragent remember	Architectural or significant code choices
convention	Post-commit hook, mindragent init	Naming and file conventions per language
bug_pattern	Post-commit hook (fix commits)	Structural fingerprint of a known bug shape
debt	Post-commit hook, mindragent debt add	TODO / FIXME / HACK markers with location
debt_resolved	mindragent debt resolve	Marker that a debt item was addressed
session_checkpoint	mindragent session checkpoint	Session summary written before context switch
note	mindragent remember	Free-form manual notes
context	Post-commit hook	Activity record for session metering

Common tags

Tags are stored as { key, value } objects in the SDK and as mindr:<key>:<value> strings on the wire (in the Remembr backend). packages/core/src/schema.ts is the conversion boundary.

Key	Values	Used on
type	See table above	All memories
module	Directory name (e.g. auth, api)	Most memories
language	typescript, python, go, rust, javascript	convention
fingerprint	SHA-256 hex string	bug_pattern
fix_commit	Git SHA	bug_pattern
severity	high, medium, low	debt
debt_id	Stable hash of file+keyword+text	debt
git_commit	Git SHA	decision, context
branch_lineage	Branch name	decision, context
reversed_decision	true	note (reversal markers)
original_decision	Memory ID	note (reversal markers)
original_debt	Memory ID	debt_resolved
source	manual	Manually-added debt and memories

Decision metadata

Decision memories carry structured metadata beyond the tag set:

{
  date: "2026-05-01",              // YYYY-MM-DD
  trigger: "keyword",              // primary trigger type
  triggers: ["keyword", "dep-change"], // all triggered signals
  confidence: 0.55,                // sum of signal weights [0, 1]
  rationale: "...",                // commit body text
  filesAffected: ["package.json", "src/api/index.ts"]
}

Debt metadata

{
  file: "src/billing/invoice.ts",
  line: 47,
  keyword: "FIXME",
  severity: "high",
  manual: true   // only present for manually-added items
}

Convention metadata

Convention memories store a full ConventionProfile in metadata:

{
  language: "typescript",
  profile: {
    language: "typescript",
    analyzedFiles: 42,
    analyzedAt: "2026-05-01T10:00:00Z",
    conventions: [
      { pattern: "camelCase", category: "functionNames", score: 97, sampleCount: 210 },
      { pattern: "PascalCase", category: "classNames", score: 100, sampleCount: 18 },
      { pattern: "kebab-case", category: "fileNames", score: 89, sampleCount: 45 }
    ]
  }
}

Wire format

When memories are stored in the Remembr backend, tags are serialised as strings in the format mindr:<key>:<value>. For example:

mindr:type:decision
mindr:module:api
mindr:git_commit:a1b2c3d4

The MindrTag ↔ wire format conversion is handled automatically by packages/core/src/schema.ts. You never need to construct wire-format strings unless writing a custom backend.