Can I turn an existing object into a Vector Store object?

No. The storage type is fixed once the object holds data. Create a new object with Vector Store storage and move the data across.

Why don't I see the Vector or Geo Point field type?

Both are gated to Vector Store objects. If the object's storage is JSON Store or anything else, only the other twelve field types appear.

Do I need a Vector Store object to use AI agents?

No. Agents read and write ordinary objects. Vector Store is only required when you want AI similarity search over the records themselves.

Which engine and algorithm should I choose?

The defaults are fine for most cases. FAISS, Lucene, and NMSLIB are the engine options and HNSW is the algorithm, with space types and EF parameters that vary per engine.

Why is "Generate via prompt" missing from Add Record?

It is gated by platform AI configuration. If platform AI is not configured, only Start from scratch, Fill via JSON, and Upload file appear.

What dimension should I set on a Vector field?

Match it to the embedding model that produced your vectors. A mismatch between the model's output size and the field's dimension will prevent the vectors from being stored.

Object Manager includes AI-specific capabilities: Vector Store objects that hold embeddings for similarity search and RAG, the Vector and Geo Point field types they unlock, and AI-assisted record creation.

Overview

Most objects are ordinary tables. Two things change when you want an object to take part in AI workflows: the storage type you pick when you create it, and the field types that storage unlocks.

Both decisions are effectively permanent. The storage type cannot be changed once the object holds data, and a field's type becomes read-only as soon as records exist. Decide before you load data.

<span class="shot">Screenshot TODO — Create object, Storage section with Vector Store selected</span>

Vector Store objects

Store data in sets the store backing the object. Vector Store is the option for embeddings — the numeric representations behind AI similarity search and RAG. The full set of choices:

JSON Store — business records with nested fields. The default, and the normal choice for structured data.
Blob Store — PDFs, images, videos, and other large files.
Key Value Store — fast key lookups for caches and sessions.
Event Store — append-only logs of activity over time.
Vector Store — embeddings for AI similarity search and RAG.
Analytics Store — time-series and aggregated metrics for dashboards.

Choose Vector Store only when the object exists to be searched by similarity. For the vast majority of objects, JSON Store is correct.

Vector fields

A Vector field stores an embedding — a list of numbers representing text or other content — so records can be matched by AI similarity search rather than exact value. This is the basis of semantic search and RAG.

Dimension: The size of the vector. Set it to match the model that produced the embeddings.
Engine: Choose FAISS, Lucene, or NMSLIB.
Algorithm: HNSW, with space types and EF parameters that vary per engine.

The defaults are fine for most cases. Vector fields appear only on objects whose storage is a Vector Store.

<span class="shot">Screenshot TODO — Vector field configuration with Dimension and engine options</span>

Geo Point fields

Geo Point stores a geographic coordinate — longitude and latitude — so you can run location queries, such as finding records near a point. Like Vector, it appears only on Vector Store objects.

Most everyday objects need neither type. Reach for them only when you are doing AI similarity search or location-based queries.

Generating records with AI

The Add Record modal asks How would you like to add your records? and offers four methods:

Start from scratch — fill a form field-by-field to create one record.
Fill via JSON — paste a JSON object or an array of records.
Upload file — import records from a CSV or Excel file. Shown if you have import permission.
Generate via prompt — auto-generate records from a description using AI.

Generate via prompt appears only when platform AI is configured.

Objects in AI agent workflows

An object's records can be read and written from automations, interfaces, and AI agents — the same object backs an app, an automation, and an agent without being redefined for each.

An ordinary JSON Store object is enough for an agent to read and write records. A Vector Store object is only needed when you want similarity search over the object's own contents.

AI capabilities on object