Skip to content

Collections

Collections are containers for organizing and managing groups of documents in PrimeThink. They support semantic search, metadata tagging, and can be shared across chats for knowledge retrieval.

Overview

A collection groups related documents together and can be attached to one or more chats. When a collection is indexed, its documents are vectorized and available for semantic search (RAG). Collections also support custom metadata via the extra field, allowing you to store arbitrary JSON data alongside the collection.

Collection Properties

Field Type Description
id number Unique collection ID
name string Collection name
description string Optional description
uuid string Collection UUID
type string Collection type (e.g., "collection")
public boolean Whether the collection is publicly accessible to the group
indexed boolean Whether the collection is indexed for semantic search
extra object|null Optional JSON object for custom metadata
external_source_type string|null External source integration type
external_source_value string|null External source reference value
tags array Array of tag objects
status string Collection status
created_at string ISO creation timestamp
last_updated_at string ISO last update timestamp

Indexing

The indexed field controls whether a collection's documents are vectorized and available for semantic search.

Behavior

  • indexed: false (default): Documents in the collection are stored but NOT searchable. Search endpoints return empty results. Document indexing tasks skip this collection.
  • indexed: true: Documents are vectorized and available for semantic search via RAG. Setting indexed to true on an existing collection triggers a reindex of all its documents.

When to Use

  • Set indexed: true when you want the collection's documents to be searchable by the AI assistant or via the search API.
  • Keep indexed: false for collections used purely as storage or when you want to control when documents become searchable (e.g., staging collections).

Search Impact

When a collection has indexed: false: - POST /collections/{id}/search returns { "indexed": false, "results": [] } - POST /collections/{id}/search/images returns { "indexed": false, "results": [] } - The collection is excluded from chat-level RAG search - Reindex requests are rejected with a 409 error

Custom Metadata (extra)

The extra field allows you to attach arbitrary JSON metadata to a collection. This is useful for: - Tracking the source or origin of the collection - Storing configuration or categorization data - Adding application-specific metadata that your Live Apps can read

Examples

{
    "extra": {
        "source": "arxiv",
        "category": "machine-learning",
        "imported_at": "2024-03-15"
    }
}

{
    "extra": {
        "department": "engineering",
        "project": "q2-roadmap",
        "priority": "high"
    }
}

The extra field is null when not set. It can be updated via the collection update endpoint.

API Endpoints

For complete API specifications, see the Interactive API Documentation.

Create Collection

curl -X POST "https://api.primethink.ai/collections?name=My%20Collection&indexed=true" \
  -H "Authorization: Token YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"extra": {"source": "manual", "category": "reports"}}'

Query Parameters: - name (string, required): Collection name - description (string, optional): Collection description - type (string, optional): Collection type (default: "collection") - public (boolean, optional): Whether collection is public (default: false) - indexed (boolean, optional): Whether to index for search (default: false)

Body (optional): - extra (object): Custom JSON metadata

Update Collection

curl -X PATCH "https://api.primethink.ai/collections/10" \
  -H "Authorization: Token YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "indexed": true,
    "extra": {"category": "updated-category"}
  }'

Setting indexed from false to true automatically triggers reindexing of all documents in the collection.

Search Collection

curl -X POST "https://api.primethink.ai/collections/10/search?query=machine%20learning" \
  -H "Authorization: Token YOUR_API_KEY"

Response when indexed: 

{
    "indexed": true,
    "results": [
        { "text": "...", "metadata": { "source": "paper.pdf", "page": 3 } }
    ]
}

Response when NOT indexed: 

{
    "indexed": false,
    "results": []
}

Using Collections in Live Pages

List Collections

const collections = await pt.listCollections();

// Filter to searchable collections
const searchable = collections.filter(c => c.indexed);

// Access custom metadata
collections.forEach(col => {
    if (col.extra) {
        console.log(`${col.name} source:`, col.extra.source);
    }
});

Search Documents in Collections

// Search documents (only works on indexed collections)
const results = await pt.searchDocuments('machine learning', 'collections');

// Search images in a specific collection
const images = await pt.searchImagesInCollection({
    collectionId: 10,
    query: 'product photos',
    topK: 5
});

Get Documents in Collections

const docs = await pt.getDocumentsInCollections([10, 15]);

// Access by collection ID
const col10Docs = docs['10'];
console.log(`Collection 10 has ${col10Docs.length} documents`);