With Embeddings

PAM embeddings are stored in a separate companion file. This keeps the core memory store lightweight while supporting systems that need vector search. This example provides embeddings for all 5 memories in the Complete Memory Store.

Example

{
  "schema": "portable-ai-memory-embeddings",
  "schema_version": "1.0",
  "embeddings": [
    {
      "id": "emb-001",
      "memory_id": "mem-001-identity",
      "model": "example-model-8d",
      "dimensions": 8,
      "created_at": "2026-02-15T22:01:00Z",
      "vector": [
        0.0123,
        -0.0456,
        0.0789,
        -0.0234,
        0.0567,
        -0.0891,
        0.0345,
        -0.0678
      ]
    },
    {
      "id": "emb-002",
      "memory_id": "mem-002-skill",
      "model": "example-model-8d",
      "dimensions": 8,
      "created_at": "2026-02-15T22:01:00Z",
      "vector": [
        0.0234,
        -0.0567,
        0.0891,
        -0.0123,
        0.0456,
        -0.0789,
        0.0678,
        -0.0345
      ]
    },
    {
      "id": "emb-003",
      "memory_id": "mem-003-project",
      "model": "example-model-8d",
      "dimensions": 8,
      "created_at": "2026-02-15T22:01:00Z",
      "vector": [
        0.0345,
        -0.0678,
        0.0123,
        -0.0456,
        0.0789,
        -0.0234,
        0.0567,
        -0.0891
      ]
    },
    {
      "id": "emb-004",
      "memory_id": "mem-004-preference",
      "model": "example-model-8d",
      "dimensions": 8,
      "created_at": "2026-02-15T22:01:00Z",
      "vector": [
        0.0456,
        -0.0789,
        0.0234,
        -0.0567,
        0.0891,
        -0.0123,
        0.0345,
        -0.0678
      ]
    },
    {
      "id": "emb-005",
      "memory_id": "mem-005-environment",
      "model": "example-model-8d",
      "dimensions": 8,
      "created_at": "2026-02-15T22:01:00Z",
      "vector": [
        0.0567,
        -0.0891,
        0.0345,
        -0.0678,
        0.0123,
        -0.0456,
        0.0789,
        -0.0234
      ]
    }
  ]
}

Note

This example uses a fictional 8-dimensional model for readability. Real embedding models typically produce 256–3072 dimensional vectors.

Key Fields

• schema / schema_version — Identifies this as a PAM embeddings file (portable-ai-memory-embeddings, version 1.0)
• embeddings[].id — Unique embedding identifier (e.g. emb-001), referenced by the embedding_ref field in the corresponding memory object in the memory store
• embeddings[].memory_id — Links back to the corresponding memory object (e.g. mem-001-identity)
• embeddings[].model — Embedding model used to generate the vector (example-model-8d in this demo)
• embeddings[].dimensions — Vector dimensionality, must match the length of the vector array (8 in this demo)
• embeddings[].vector — The actual embedding vector as an array of floats
• embeddings[].created_at — When the embedding was generated; useful for detecting stale embeddings after memory content is updated

The embeddings file validates against portable-ai-memory-embeddings.schema.json.