Weaviate

Open-source vector database for agent memory. Manage collections and objects, and run vector, semantic, keyword, and hybrid searches, with a raw GraphQL escape hatch for advanced retrieval.

20 actions available

A batch of new records is ready to enter the collection

Agent reads each record and its vector from an upstream step

Agent writes with Batch Create Objects for speed

Agent confirms the count with Aggregate Count

Agent runs Search (Hybrid) to confirm the content is retrievable

Agent reports the collection size change to the owning team

Any Batch Delete Objects run waits for an owner to approve

What This Integration Enables

The Weaviate connector covers collections, objects, search, and utilities. Agents create and manage collections (classes) with typed properties and optional vectorizer modules, and create, get, update, replace, delete, and list objects, plus batch create and batch delete. The search actions cover Search (Vector), Search (Text), Search (Keyword) with BM25, and Search (Hybrid), which blends vector and keyword ranking on an alpha parameter. GraphQL Query (Raw) is the escape hatch for features the dedicated actions do not cover, such as groupBy, sort, and cross-reference resolution. Get Meta and Check Liveness report on the instance. Search (Text) and Search (Hybrid) rely on the collection's vectorizer module, so an agent supplies the module API key through configuration when using an external provider. Agents commonly vectorize content with a provider such as [OpenAI](/integrations/openai-ai) before a Batch Create Objects call, then hand retrieved records to an LLM or log them into a system like [Google Sheets](/integrations/google-sheets).

Without FlowRunner

Search infra per project Vector, keyword, and hybrid search each get built from scratch

Slow one-by-one sync Records are inserted individually, one call at a time

Blind cleanups A filter delete removes matches nobody previewed

With FlowRunner

Four search modes Vector, text, keyword, and hybrid search are all single actions

Bulk sync Batch Create Objects indexes many records in one request

Previewed cleanups A dry run shows what a batch delete would remove first

Use Case Scenarios

Hybrid Search Backend

A team wants search that catches both exact terms and semantic matches. The agent indexes documents with Batch Create Objects into a collection that uses a vectorizer module. On a user question it calls Search (Hybrid) with an alpha that leans toward vector recall, so a query for a concept finds the right passages even when the wording differs. Keyword and vector recall combine in one call, and the top records feed an LLM.

Bulk Sync From Another System

Records live in an external source and change often. On a schedule the agent pulls the changed records, embeds them, and writes them with Batch Create Objects, which is far faster than creating them one at a time. Aggregate Count confirms the collection size after the run. The vector store tracks the source without a per-record round trip.

Cleaning Out Stale Objects

Old objects should be removed on a policy, for example anything older than a retention window. The agent first runs Batch Delete Objects with Dry Run enabled to preview how many objects the where filter would remove, assembles that count into a summary, and routes it to the data owner. Only after approval does it run the delete for real.

Human-in-Loop Highlight

Searching a collection is read-only. A batch delete removes every object a where filter matches, up to the server limit per call. [Human-in-the-loop](/concepts/human-in-the-loop/) is an execution pattern where AI agents pause autonomously, assemble the relevant context and the decision choices available, route to a human via their preferred channel, and resume the moment the human responds. The Weaviate connector places that pause on Batch Delete Objects, and it has exactly the right tool to make the pause meaningful: Dry Run. When a flow reaches a batch delete, the agent runs it as a dry run first to get the exact match count without deleting anything, then asks the owner through their channel: "This retention cleanup would delete 8,700 objects from the Article collection matched by the archived filter. Run it for real?" The agent handles indexing, sync, and search on its own. A person confirms the delete after seeing precisely what it would remove.

Agent processes routinely

Detects exception requiring judgment

Clear match Continues automatically

Ambiguous Routes to human via preferred channel

Human decides

Agent resumes with decision

Agent Capabilities

20 actions

Collections

Create Collection Create a new collection (class) in the Weaviate schema, optionally with a vectorizer module, typed properties, and advanced configuration.
List Collections Retrieve the full Weaviate schema, including every collection with its properties, vectorizer, and module configuration.
Get Collection Retrieve the schema definition of a single collection, including its properties, vectorizer, and index configuration.
Delete Collection Permanently delete a collection from the schema, including all objects stored in it. This cannot be undone.

Objects

Create Object Create a single data object in a collection. The vector is generated automatically when the collection uses a vectorizer, or supplied explicitly.
Get Object Retrieve a single data object by its collection and UUID, optionally including its stored embedding vector.
Update Object (Merge) Partially update an existing object by merging the provided properties into it. Properties not included keep their current values.
Replace Object Replace an existing object entirely. All previous properties are discarded and replaced with the provided ones.
Delete Object Permanently delete a single data object by its collection and UUID. This cannot be undone.
List Objects List data objects in a collection with cursor-based pagination using an After cursor.
Batch Create Objects Create multiple data objects in a single batch request, much faster than one by one. Returns per-object success or failure status.
Batch Delete Objects Delete all objects in a collection that match a where filter, in a single batch request, with an optional dry run to preview the count.

Search

Search (Vector) Perform a vector similarity search (nearVector) using an embedding you supply. Works with any collection, including those without a vectorizer.
Search (Text) Perform a semantic text search (nearText). Requires the collection to use a vectorizer module, such as text2vec-openai.
Search (Keyword) Perform a keyword search using the BM25 ranking algorithm. Works on any collection without a vectorizer.
Search (Hybrid) Perform a hybrid search combining vector similarity and BM25 keyword ranking, with an alpha parameter to balance the two.
GraphQL Query (Raw) Execute an arbitrary GraphQL query against the Weaviate endpoint: an escape hatch for advanced features not covered by the dedicated search operations.
Aggregate Count Count the objects in a collection using a GraphQL Aggregate query, optionally restricted by a where filter.

Utilities

Get Meta Retrieve metadata about the Weaviate instance, including server version, hostname, and enabled modules.
Check Liveness Check whether the Weaviate instance is alive and ready to receive traffic via the readiness endpoint.

Start building with Weaviate

$100 in credits. No card required. Connect in minutes.

Start Building Free Talk to us