Adding a Store

Implement StoreAdapter to add support for a new database.

Interface

class StoreAdapter {
  async initialize() {}
  async createMemory(memory) {}
  async getMemory(id) {}
  async queryMemories(filters, options) {}
  async vectorSearch(embedding, options) {}
  async updateMemory(id, patch) {}
  async deleteMemory(id) {}
  async createRelation(relation) {}
  async getMemoryGraph(memoryId, options) {}
  async health() {}
  async getMigrationVersion() {}
  async runMigration(sql) {}
  static get metadata() {}
}

Schema requirements

Your store must support the Memory record shape:

{
  id: string,           // e.g., "mem_abc123"
  rawRef: string,       // blob reference
  contentType: string,  // MIME type
  extractedText?: string,
  summary?: string,
  embedding?: number[],
  metadata?: object,
  parentId?: string,
  sector?: string,      // rich mode
  salience?: number,    // rich mode
  validFrom?: Date,     // rich mode
  validTo?: Date,       // rich mode
  access?: object,      // rich mode: { owner, readers[], writers[], policy? }
  createdAt: Date,
  updatedAt?: Date
}

Vector search

vectorSearch(embedding, options) must return results ranked by similarity:

[
  { memory: MemoryRecord, score: 0.94 },
  { memory: MemoryRecord, score: 0.89 }
]

Use your database's vector extension (pgvector, sqlite-vec, etc.) or implement cosine similarity in application code.

Relations and graph

createRelation and getMemoryGraph power the knowledge graph:

// createRelation
{
  sourceId: "mem_abc",
  targetId: "mem_def",
  type: "supersedes",
  confidence: 0.95,
  metadata: { reason: "updated belief" }
}
 
// getMemoryGraph returns related memories with path info

Migrations

Your adapter must manage its own schema. On initialize(), check getMigrationVersion() and run any pending migrations via runMigration().

Testing

Run the store contract test:

npm run test:integration:store -- --provider=qdrant