Knowledge Base

Knowledge Base components connect the flow to semantic search indexes for retrieval-augmented generation.

Knowledge Retrieval

Retrieves relevant chunks from a knowledge base using semantic search. Connects to knowledge bases configured in the Foundry and can search across multiple bases in one run.

Use in a flow

  1. Add a Knowledge Retrieval node to the canvas.

  2. Select one or more Knowledge Bases from the workspace dropdown (refresh to load the latest list).

  3. Provide a Search Query (wire from Input, Prompt, or another upstream component).

  4. Connect Search Results or DataFrame to downstream steps (for example, Prompt or an Agent).

Input (Message) ──► Knowledge Retrieval ──► Prompt ──► Agent

Results from all selected knowledge bases are aggregated and sorted by relevance score.

Parameters

Parameter

Default

Hidden

Description

Knowledge Bases

—

No

One or more knowledge bases to search. Refresh to reload options. Required.

Search Query

—

No

The question or text to search for. Required.

Fallback Knowledge Base

—

Yes

Knowledge base used when the user lacks access to a selected primary KB. Required.

Top K Results

4

Yes

Maximum number of results to return per knowledge base.

Search Type

similarity

Yes

similarity (vector search) or hybrid_search (combines keyword and vector search).

Search Score Threshold

0.0

Yes

Minimum similarity score (0.0–1.0). Results below this threshold are excluded.

Search Metadata

—

Yes

JSON metadata filters to narrow results (e.g., {"department": "HR"}).

Use SSL

true

Yes

Use SSL when connecting to OpenSearch.

Verify Certificates

false

Yes

Verify SSL certificates for the OpenSearch connection.

Save Output as Variable

false

Yes

Store this component’s output as a named flow variable.

Variable Name

—

No

Name for the flow state variable when Save Output as Variable is enabled.

Outputs

Output

Description

Search Results

Matching document chunks as Data objects (text and metadata).

DataFrame

Search results as a table.

Limitations

  • Foundry configuration — Knowledge bases must exist in the workspace and be accessible to the current user.

  • Search query required — Hybrid search requires a non-empty query.

  • Multi-KB search — Top K applies per knowledge base, not globally across all selected bases.

  • KB types — Supports in-house and Bedrock knowledge bases. Bedrock backends use the Bedrock Retrieve API internally.

OpenSearch

Queries an Amazon OpenSearch index directly with configurable search type, scoring threshold, and hybrid search support.

Use in a flow

  1. Add an OpenSearch node to the canvas.

  2. Connect an Embedding model component.

  3. Select an Index Name (knowledge base index).

  4. Provide a Search Query, or connect Ingest Data to add documents before searching.

  5. Connect Search Results or DataFrame downstream.

Embeddings ──► OpenSearch ◄── Split Text (Ingest Data)
Input ──► OpenSearch (Search Query) ──► Prompt ──► Agent

Parameters — Connections and search

Parameter

Default

Hidden

Description

Index Name

—

No

OpenSearch index to query. Select from configured knowledge base indexes. Required.

Embedding

—

No

Embedding model used to vectorize the search query. Required.

Ingest Data

—

No

Data or DataFrame to add to the index before searching.

Search Query

—

No

Query text for similarity search.

Search Type

similarity

Yes

similarity, similarity_score_threshold, or mmr (max marginal relevance).

Number of Results

4

Yes

Maximum number of results to return.

Search Score Threshold

0.0

Yes

Minimum score when Search Type is similarity_score_threshold.

Hybrid Search Query

—

Yes

Custom OpenSearch hybrid query in JSON format. When set, overrides standard search.

Search Metadata

—

Yes

JSON metadata filters to narrow results.

Parameters — Connection settings

Parameter

Default

Hidden

Description

OpenSearch URL

—

Yes

Auto-configured from environment based on index type.

Username

admin

Yes

OpenSearch username.

Use SSL

true

Yes

Use SSL for the OpenSearch connection.

Verify Certificates

false

Yes

Verify SSL certificates.

Cache Vector Store

true

Yes

Reuse the same vector store instance within a single flow run.

Enable Retry

false

Yes

Retry failed search requests.

Max Retries

3

Yes

Maximum retry attempts.

Outputs

Output

Description

Search Results

Matching document chunks as Data objects.

DataFrame

Search results as a table.

Ingested Data

Documents that were ingested into the index during this run.

Limitations

  • Embedding required — A connected Embeddings component is required for vector search.

  • Index selection — Index Name must match a configured knowledge base index in the workspace.

  • Hybrid search — Hybrid Search Query must be valid OpenSearch JSON; invalid JSON causes an error.

  • Direct OpenSearch — Connection details are resolved from platform configuration based on the selected index type.