Image Search Block

The Image Search block indexes and searches images and DWG/DXF CAD drawings using keyword, visual similarity, hybrid, and spatial search. Reach for it when you need to build a searchable catalog from many file sources (upload, network, S3, Azure Blob, Google Drive, databases, or URLs) and query it at scale.

Overview

When to Use

• Build a searchable catalog of CAD drawings (DWG/DXF) and images that scales to 100K+ files.
• Run full-text keyword search across extracted text, dimensions, and part names.
• Find visually similar images or drawings from a reference image URL.
• Perform spatial search filtered by CAD layer or block name (e.g. DIMENSIONS, TITLE_BLOCK).
• Ingest files from diverse sources such as network folders, S3, Azure Blob, Google Drive, PostgreSQL/MSSQL, or URLs.
• Check indexing progress and file counts for a catalog before querying it.

Configuration

Operation

Dropdown selecting the action to perform. Defaults to Search.

Create Catalog fields

Shown when Operation = image_search_catalog.

Catalog Name (name, required) — Human-readable name for the new catalog.

Description (description) — Optional description of what the catalog contains.

Processing Mode (processingMode) — How files are processed after ingestion.

Extraction Method (extractionMethod) — How text and metadata are extracted from files. Defaults to Auto Detect.

Ingest Files fields

Shown when Operation = image_search_ingest.

Catalog ID (catalogId, required) — ID of the existing catalog to ingest files into.

Data Source (dataSource) — Where files are sourced from. Defaults to File Upload. Each choice reveals its own connection fields.

Network / Local Folder fields (visible when dataSource = network):

• networkPath — Absolute path to the folder to scan (e.g. /shared/drawings or \\server\share\drawings).
• networkRecursive (switch) — Enable to scan subfolders recursively.

AWS S3 fields (visible when dataSource = s3):

• s3Bucket — S3 bucket name.
• s3Prefix — Key prefix / folder path inside the bucket.
• s3Region — AWS region (e.g. us-east-1).
• s3AccessKey — AWS access key ID (stored as a secret).
• s3SecretKey — AWS secret access key (stored as a secret).

Azure Blob Storage fields (visible when dataSource = azure_blob):

• azureConnectionString — Full Azure storage connection string (stored as a secret).
• azureContainer — Container name.

Google Drive fields (visible when dataSource = google_drive):

• googleDriveFolderId — Google Drive folder ID.
• googleDriveAccessToken — OAuth access token (stored as a secret).

PostgreSQL fields (visible when dataSource = postgresql):

• dbConnectionString — PostgreSQL connection string, e.g. postgresql://user:pass@host:5432/dbname (stored as a secret).
• dbQuery — SQL query whose result rows contain file paths.

MSSQL fields (visible when dataSource = mssql):

• mssqlConnectionString — MSSQL connection string (stored as a secret).
• mssqlQuery — SQL query whose result rows contain file paths.

URL fields (visible when dataSource = url):

• urls — Comma-separated list of file URLs to ingest.

Extraction Method (ingestExtractionMethod) — Extraction method for this ingest run (same options as Create Catalog). Defaults to Auto Detect.

Search fields

Shown when Operation = image_search_search.

Catalog ID (searchCatalogId, required) — ID of the catalog to search.

Search Query (query) — Text to search for (text, dimensions, part names, etc.).

Search Mode (mode) — Algorithm used to find results. Defaults to Hybrid.

Reference Image URL (imageUrl) — URL of a reference image used when mode is visual.

Results Count (topK, slider 1–100) — Maximum number of results to return.

CAD Layer Filter (spatialLayer) — Filter results to drawings that contain a specific CAD layer name. Visible when mode is spatial.

CAD Block Filter (spatialBlock) — Filter results to drawings that contain a specific CAD block name. Visible when mode is spatial.

Status fields

Shown when Operation = image_search_status.

Catalog ID (statusCatalogId, required) — ID of the catalog to check.

API Key (all operations)

apiKey — OpenAI API key used for embedding generation and AI Vision (GPT-4o) extraction. Applies to all operations that need it.

Inputs & Outputs

• Inputs: operation (string) — Operation to perform; apiKey (string) — OpenAI API key; name (string) — Catalog name; description (string) — Catalog description; processingMode (string) — batch or realtime; extractionMethod (string) — Extraction method; catalogId (string) — Catalog ID; dataSource (string) — Data source type; networkPath (string) — Network folder path; networkRecursive (boolean) — Scan subfolders; s3Bucket (string) — S3 bucket name; s3Prefix (string) — S3 key prefix; s3Region (string) — AWS region; s3AccessKey (string) — AWS access key; s3SecretKey (string) — AWS secret key; azureConnectionString (string) — Azure connection string; azureContainer (string) — Azure container name; googleDriveFolderId (string) — Google Drive folder ID; googleDriveAccessToken (string) — Google Drive access token; dbConnectionString (string) — PostgreSQL connection string; dbQuery (string) — PostgreSQL SQL query for file paths; mssqlConnectionString (string) — MSSQL connection string; mssqlQuery (string) — MSSQL SQL query for file paths; urls (string) — Comma-separated file URLs; ingestExtractionMethod (string) — Extraction method for ingestion; searchCatalogId (string) — Search catalog ID; query (string) — Search query; mode (string) — Search mode; imageUrl (string) — Reference image URL; topK (number) — Number of results; spatialLayer (string) — CAD layer filter; spatialBlock (string) — CAD block filter; statusCatalogId (string) — Status check catalog ID
• Outputs: results (json) — Search results with file paths and extracted text; totalResults (number) — Total results count; catalogId (string) — Created/queried catalog ID; message (string) — Operation result message; processed (number) — Number of files processed; failed (number) — Number of files failed; status (string) — Catalog status; indexingProgress (number) — Indexing progress (0-100); fileCount (number) — Total indexed files

Tools

• Image Search Create Catalog (image_search_catalog) — Creates a new named image catalog with a chosen processing mode and extraction method. Returns the new catalogId and a confirmation message.
• Image Search Ingest (image_search_ingest) — Discovers and indexes image/CAD files from a specified data source (upload, network, S3, Azure Blob, Google Drive, PostgreSQL, MSSQL, or URL) into an existing catalog. Returns counts of discovered, processed, and failed files.
• Image Search (image_search_search) — Searches a catalog using keyword full-text, visual similarity, hybrid, or spatial (CAD layer/block) modes. Accepts a text query and/or a reference image URL. Returns ranked results with file paths, extracted content, and similarity scores.
• Image Search Status (image_search_status) — Checks the current status and indexing progress of a catalog (active, indexing, or error) along with file count, document count, and the last indexed timestamp.

YAML Example

image_search_1:
  type: image_search
  name: "Image Search"
  inputs:
    operation: "image_search_search"
    searchCatalogId: "{{start.catalogId}}"
    query: "title block revision A"
    mode: "hybrid"
    topK: 10
    apiKey: "{{OPENAI_API_KEY}}"
  connections:
    outgoing:
      - target: next-block-id