Vector Search

Find records by what they mean, not the exact words

Semantic search for your business data. Find the right record from a rough description, catch duplicates before they are saved, and surface related records automatically.

Why it matters

Keyword search only finds what you can spell exactly.

People don't remember exact words

Search “client unhappy with delivery” and find the ticket titled “late shipment complaint”.

Duplicates cost real money

Two records for the same customer means double outreach and wrong reports. Catch them when they are typed in.

Useful context is scattered

Related articles, past cases and similar orders surface on a record, even when nobody linked them.

How it works

Set it up once per entity. It stays up to date on its own.

1

Turn it on for an entity

In Admin, choose the entity and the text fields that describe a record, such as name, subject and notes.

2

Scratchboard indexes your records

Each record is converted into an embedding: a numeric fingerprint of its meaning. New and edited records are indexed automatically.

3

Search and match by meaning

Use AI search in lists and the global search, see similar records, and get duplicate warnings when someone creates a record.

Try a search by meaning

Pick an example query. Results are ranked by how close their meaning is, not by shared words.

AWS Infrastructure Migration Plan

Moving legacy systems to AWS cloud...

94%

Cloud Architecture Review

Evaluating cloud-native deployment...

89%

Enterprise SaaS Platform Setup

Setting up scalable cloud infrastructure...

82%

Data Center Decommission Project

Transitioning from on-premise to cloud...

78%

DevOps Pipeline Modernization

Implementing CI/CD with cloud services...

71%

Catch duplicates as they are typed

When someone creates a record, likely duplicates appear first, even with different spellings.

New Record

Acme Corporation
john@acme.com
2 potential duplicates found

Acme Corp

Email: john.smith@acme.com

Status: Active · Created: Jan 2024

94%

ACME Corporation Ltd.

Email: contact@acme.com

Status: Active · Created: Mar 2023

87%

Surface related records

A record page can suggest related records from other entities, with no link set up by hand.

Source Record

Enterprise Cloud Migration

Entity: Proposals

Knowledge Base3 matches

Cloud Migration Best Practices

89%

AWS vs Azure Comparison

82%

Security Considerations for Cloud

76%
Case Studies2 matches

Acme Corp Cloud Transition

85%

GlobalTech Infrastructure Upgrade

79%
Emails1 match

RE: Cloud migration timeline

72%

Key capabilities

Everything included with an index.

AI search

Switch any indexed list or the global search to search by meaning.

Duplicate warnings

See likely duplicates before saving, and review clusters to merge.

Find similar

Open the records most similar to the one you are viewing.

Discovered connections

Related records from other entities suggested on the record page.

Unusual records

A badge flags records that look very different from the rest.

In workflows

Search, find similar and embed steps you can use in any workflow.

Per-entity setup

Choose which entities and fields are indexed, and check index health.

Proven embeddings

Meaning is captured with Voyage AI or OpenAI embedding models.

Frequently asked questions

Short answers to common questions.

What is vector (semantic) search?

Normal search matches the exact words you type. Vector search compares meaning, so “refund request” also finds “customer wants money back”.

Do I have to index every entity?

No. You turn it on per entity and choose which fields are indexed. Entities without an index keep using normal text search.

Which AI provider does it use?

Embeddings are created with Voyage AI or OpenAI embedding models.

Does it replace normal search?

No. Text search still works everywhere. On indexed entities you can switch between text and AI search.

Related features

These work hand in hand with this one.

Build your first app today.

Start your 7-day free trial, or talk to us about your use case.

See pricingBrowse use casesDevelopers