EDB Docs - EDB Postgres AI v1.4.0 (LTS)

The EDB Knowledge Base component runs a semantic similarity search against an existing AIDB knowledge base in a Hybrid Manager (HM) cluster, with optional structured filtering on the knowledge base's source-table columns.

Choosing the right component

Use the EDB Knowledge Base component when:

You're building a RAG flow that retrieves relevant passages by similarity before passing them to a generation step.
You want an agent tool that looks up information in a knowledge base on demand. The component is tool-mode compatible.
You want hybrid search: semantic ranking combined with a structured filter on the knowledge base's source-table columns.

Use a different component if:

You need to manage pgvector index structure (create, drop, rebuild HNSW or IVFFlat indexes) on an existing vector column. Use EDB Vector Index.
You need direct SQL access to a database, not similarity search. Use EDB Database.
The knowledge base doesn't exist yet. Create one through Pipeline Designer or by calling aidb.create_table_knowledge_base() against the cluster.

Prerequisites

The Project Viewer role on the project that owns the cluster:
- Project > Users > Assign Project Roles > Project Viewer.
An HM cluster with at least one AIDB knowledge base. Create one through Pipeline Designer or directly via aidb.create_table_knowledge_base(). The knowledge base's source can be a table (filtering supported) or a volume (semantic-only).
An HM machine-user access key and the database user/password saved in Langflow as Global Variables.

Inputs

Connection

Field	Type	Required	Default	Notes
Connection Mode	Dropdown	Yes	Hybrid Manager	`Hybrid Manager` or `Direct Connection`. Switches the connection fields that follow.
Hybrid Manager URL	Text	No		Advanced. Override the default HM URL.
HM Machine User Key	Secret	Yes	`HM_API_KEY`	Defaults to the global variable named `HM_API_KEY`.
Hybrid Manager Project	Dropdown	Yes		Populated from your HM projects.
Hybrid Manager Database	Dropdown	Yes		The cluster containing the knowledge base.
Database Group	Dropdown	No		Cluster group, for example primary or reader.
Database Connection Type	Dropdown	Yes		For example, read or read/write.
User for the database	Text	Yes	`HM_DB_USER`	Postgres role. Defaults to the `HM_DB_USER` global variable.
Password for the database	Secret	Yes	`HM_DB_PASSWORD`	Defaults to the `HM_DB_PASSWORD` global variable.
Default database name override	Text	No		Advanced. Override the cluster default when listing databases.
Database Name	Dropdown	Yes		The database holding the AIDB knowledge bases.

In Direct Connection mode the HM cluster fields are replaced with Database Host and Database Port. The user, password, and name fields remain.

Knowledge base and query

Field	Type	Required	Default	Notes
Knowledge Base	Dropdown	Yes		Populated from `aidb.knowledge_bases` in the selected database. Use the refresh button if you've just created one.
Limit results	Integer	Yes	`10`	Maximum rows to return.
Query Value	Multiline text	Yes		The natural-language query to search for. Tool-mode compatible: agents call the component with this as the tool argument.

Search mode

Field	Type	Required	Default	Notes
Search Mode	Dropdown	Yes	`Vector Only`	`Vector Only` runs pure semantic search via `aidb.retrieve_*`. `Vector + Filter` adds a `WHERE` clause against the knowledge base's source table. Only `Vector Only` is offered when the KB's source is a volume. See Hybrid search.

Filter (Vector + Filter only)

These fields appear only when Search Mode is Vector + Filter and the knowledge base has a table source.

Field	Type	Required	Default	Notes
Filter Column	Dropdown	Yes		Populated from the source table's columns.
Operator	Dropdown	Yes	`=`	See Filter operators.
Filter Value	Text	Yes		Value to compare against. For `IN`, supply a comma-separated list. Ignored for `IS NULL` / `IS NOT NULL`.

Output options

Field	Type	Required	Default	Notes
Include Text	Boolean	No	`false`	Applies in Vector Only mode: when on, the result includes the source-text column for each hit (uses `aidb.retrieve_text`); when off, only keys and distances are returned (`aidb.retrieve_key`). Has no effect under Vector + Filter, where the source-text column is always returned.
Include Columns	Boolean	No	`true`	Advanced. Include source-table column metadata in the output. Tool-mode compatible.
Add Error	Boolean	No	`false`	Advanced. Surface query errors in the result instead of raising. Tool-mode compatible.

Outputs

Output	Type	Carries
Result Table	DataFrame	One row per match, with `key`, optional `value` (text), and `distance`. Distance is computed using the knowledge base's configured metric (Cosine, L2, or Inner Product).

Hybrid search

Vector + Filter combines semantic similarity with a structured WHERE clause against the knowledge base's source table. Use it when you want the top-K semantic matches within a subset of your data. For example, "the closest match in documents tagged kind = 'policy'".

How it works:

The component JOINs the KB's vector table with its source table on the source key column.
The filter is applied in the WHERE clause before the order-by-distance.
For HNSW-indexed vectors the component sets hnsw.iterative_scan = relaxed_order so the filtered search still returns useful results when the index would otherwise prune too aggressively.

Hybrid search is only available when the knowledge base's source is a table. If the source is a volume, the Search Mode dropdown is restricted to Vector Only and the filter fields stay hidden.

Filter operators

Operator	Behavior
`=`, `!=`, `>`, `<`, `>=`, `<=`	Standard SQL comparison. Numeric values are passed unquoted. Non-numeric values are quoted.
`LIKE`	SQL `LIKE` pattern match. Use `%` wildcards in Filter Value.
`IN`	Comma-separated values in Filter Value, for example `policy, contract, memo`.
`IS NULL`, `IS NOT NULL`	Filter Value is ignored.

EDB Knowledge Base v1.4.0 (LTS)