Runtype
Records

What are Records?

Records are structured data entries with JSON metadata that Flows and Agents use as context, knowledge bases, or data storage.

Use cases for Records

  • Knowledge bases: Store documentation, FAQs, or Product information for AI retrieval

  • Customer data: Save customer profiles, preferences, order history

  • Product catalogs: Maintain inventory, specifications, pricing

  • Conversation history: Store chat logs or interaction summaries

  • RAG (Retrieval Augmented Generation): Provide context to AI responses

Record structure

Every Record has:

  • ID: Unique identifier (auto-generated by Runtype in TypeID format with prefix record, e.g. record_01h2x...)

  • Name: Required display label (max 500 characters, e.g. record title or description)

  • Type: Category or collection (e.g., "customers", "products")

  • Metadata: JSON object with your data

  • Embeddings: Vector representations for semantic search (added via Flow steps)

Example Record:

{
  "id": "record_01h2xabc123...",
  "name": "Jane Doe",
  "type": "customers",
  "metadata": {
    "email": "[email protected]",
    "tier": "premium",
    "totalOrders": 47,
    "lastPurchase": "2024-01-10"
  }
}

Metadata flexibility

Metadata can be any valid JSON. Structure it to match your needs:

{
  "productId": "PROD-123",
  "name": "Widget Pro",
  "price": 99.99,
  "specs": {
    "weight": "2.5kg",
    "dimensions": "10x5x3cm"
  },
  "tags": ["electronics", "popular"]
}

The platform also supports metadata labels—user-defined display names for metadata keys. These labels appear in the dashboard Records list and table headers, making it easier to read and export your data.

Semantic search with embeddings

To use semantic search, add generate-embedding and store-vector steps to your Flow. These steps generate vector embeddings from Record fields and save them to the Record, enabling semantic search that finds Records by meaning, not exact keywords.

Example: Search "affordable laptop for students" finds Records about budget laptops, even if they don't contain those exact words.

Record types

Organize Records into types (like database tables):

  • customers — Customer profiles

  • orders — Order data

  • docs — Documentation articles

  • faqs — FAQ content

Each type can have different metadata structures.

Managing Records in the dashboard

The Records page lists all your Records with columns for Name, Type, Metadata keys (first three shown as badges, using metadata labels if set), Created, and Updated. You can filter by type and search by name or metadata text. Bulk actions let you select and delete multiple Records at once.

Use Records for data that changes infrequently or needs to be queried by AI. For dynamic data updated constantly, use external databases and API calls.

Next steps

Was this helpful?