Search Engine - Cloosphere Guide

Admin › Settings › Search Engine

Set the vector database type and search parameters.

These settings are shared across Knowledge Bases (KbSphere), Glossaries, and DbSphere.

Search Engine Selection

Engine	Strengths
Azure AI Search	Azure managed service, hybrid search supported by default
PostgreSQL pgvector	PostgreSQL extension, memory-efficient halfvec support
Milvus	Large-scale distributed processing, high-performance vector search
Elasticsearch	Hybrid search (BM25 + vector), reuse existing ELK stack
Google Vertex AI Search	Google Cloud managed search service

Azure AI Search
pgvector
Milvus
Elasticsearch
Vertex AI Search

Setting	Description
Endpoint	Azure Search service URL
API Key	Admin API Key
API Version	API version (e.g., `2024-07-01`)

Setting	Description
Host	PostgreSQL server address
Port	Port number
Database	Database name
User	Connection username
Password	Connection password

The pgvector extension must be installed in PostgreSQL.

Setting	Description
URI	Milvus server address
Token	Auth token (optional)
User	Auth username (optional)
Password	Auth password (optional)

Setting	Description
URL	Elasticsearch server URL
Username	Auth username (optional)
Password	Auth password (optional)
Index Prefix	Index name prefix (optional)

Setting	Description
Project ID	Google Cloud project ID
Location	Region
Collection	Data store collection
Use Global Google Cloud Key	When enabled, uses the global Google Cloud key. When disabled, enter Service Account Key JSON directly below.
Service Account Key JSON	JSON input field shown when `Use Global Google Cloud Key` is disabled. Paste the full service account key JSON.

Embedding engines and models are configured in the Settings > Documents tab. Vector dimensions must match between search engine and embedding settings.

Search Settings

Configure the number of results to retrieve from vector search.

Setting	Description	Default
Top K	Number of search results to retrieve from the vector database	`10`

Adjusting Top K:

Higher — retrieves more candidate documents and lowers the chance of missing the answer, but increases irrelevant results (noise), context tokens, and cost, and slows responses.
Lower — precise, fast, and token-saving, but may miss the document containing the answer.

Reranker

The reranker re-ranks search results with a semantic model to improve accuracy (optional).

Select the reranker to use from the Reranker Type dropdown and configure the related parameters.

Setting	Description	Default
Reranker Type	Type of reranker to use	`Azure AI Search (built-in)`
Reranker Top K	Result count after reranker	`5`
Reranker threshold	Minimum relevance score after reranking (0–1). Results below threshold are filtered	`0.0`

Tuning guide — Reranker Top K decides how many of the results fetched by Top K are finally used, so always keep it at or below Top K.

Reranker Top K — Raising it passes more evidence to the LLM and reduces omissions but increases noise and tokens; lowering it keeps only the most relevant few for precision but may miss results.
Reranker threshold — Raising it filters out low-relevance results and reduces noise, but setting it too high lets no results through and returns an empty search. 0 passes everything with no filter.

When you select Vertex AI Ranking API as the reranker type, the sub-fields below are shown.

Setting	Description
Project ID	Google Cloud project ID (optional — uses Vertex AI Search’s Project ID when unset)
Location	Vertex AI Ranking API region
Reranker Model	Reranking model to use
Authentication	Auth method (Use Global Google Cloud Key toggle or Service Account Key JSON entry)

Changing the search engine type makes previously indexed data inaccessible. Always plan data migration before changing.

​Search Engine Selection

​Search Settings

​Reranker

Search Engine Selection

Search Settings

Reranker