Qdrant

Open-source vector database and similarity search engine for agent memory. Manage collections, upsert and retrieve points with vectors and payloads, run similarity searches, and maintain point metadata.

16 actions available

New content is ready to enter the collection

Agent reads the point vector and payload from an upstream step

Agent writes with Upsert Points, waiting for the change to apply

Agent confirms the change with Count Points

Agent runs Query Points to confirm the content is retrievable

Agent reports the collection size change to the owning team

Any filter-based Delete Points or Overwrite Payload waits for approval

What This Integration Enables

The Qdrant connector spans collections, points, and payload. Agents create, list, get, delete, and check collections, then upsert, get, delete, query, batch-query, scroll, and count points. Query Points is the universal similarity search: query by a raw embedding or by an existing point ID, with payload filtering, score thresholds, and named vectors. Point IDs can be unsigned integers or UUIDs, and payload operations, Set, Overwrite, Delete Keys, and Clear, select points either by an explicit ID list or by a Qdrant filter. Write operations wait until the change is applied before returning, so a downstream step reads a consistent store. The connector unwraps the REST envelope and returns the result directly. Agents commonly embed content with a provider such as [OpenAI](/integrations/openai-ai) or [Gemini AI](/integrations/gemini-ai) before an Upsert Points call, then surface top matches to a channel like [Slack](/integrations/slack).

Without FlowRunner

Metadata lives apart Vectors sit in one store and their context in another

Store falls behind Points drift from the source as content changes

Blind bulk edits A filtered update touches whatever it matches, seen or not

With FlowRunner

Vectors with payload Each point carries its metadata, filterable at query time

Store stays current Upsert Points keeps vectors and payloads in step with source

Reviewed bulk edits A person confirms filter-based writes before they run

Use Case Scenarios

RAG With Rich Payloads

A team stores documents as points, each carrying source, author, and date in its payload. The agent embeds each chunk and writes it with Upsert Points, attaching the metadata. On a user question it runs Query Points with a payload filter that scopes to recent documents from an approved source, then hands the top passages to an LLM. The payload turns retrieval into something an agent can constrain, not just rank.

Paging Through A Large Collection

An agent needs to process every point that matches a condition. It uses Scroll Points to page through the collection with a filter, carrying the returned next-page offset forward until the results run out. Because Scroll Points does not score by similarity, it is the right tool for a full sweep rather than a top-K query.

Reclassifying A Segment

A metadata field needs to change across many points, for example retagging a batch of documents. The agent identifies the affected points with a filter and Count Points, assembles a summary of how many payloads a change would touch, and routes it to the data owner. Only after approval does it run Set Payload or Overwrite Payload against the filter.

Human-in-Loop Highlight

Querying points is read-only. A filter-based write can rewrite thousands of points the agent never inspected one by one. [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 Qdrant connector places that pause on the operations that act by filter rather than by explicit ID: a Delete Points that selects by filter, or an Overwrite Payload that replaces metadata across a matched set. When a flow reaches such a step, the agent first counts the matches with Count Points, then asks the owner through their channel: "This overwrite would replace the payload on 3,900 points matched by the archived filter. Previous payload fields are removed. Proceed?" The agent handles upserts and single-point edits on its own. A person owns the bulk write whose blast radius is a filter, not a list.

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

16 actions

Collections

Check Collection Exists Check whether a collection with the given name exists in the Qdrant instance.
Create Collection Create a new vector collection with a given vector size and distance metric, with an optional advanced configuration for named vectors, quantization, and HNSW tuning.
Delete Collection Permanently delete a collection and all points stored in it. This cannot be undone.
Get Collection Retrieve detailed information about a collection, including status, vector configuration, and points count.
List Collections Retrieve the names of all collections that exist in the Qdrant instance.

Points

Batch Query Points Execute multiple similarity-search queries against a collection in a single request, returning one result set per search.
Count Points Count the points in a collection, optionally restricted to those matching a filter.
Delete Points Delete points from a collection by an explicit list of point IDs or by a Qdrant filter. Provide exactly one selector.
Get Points Retrieve specific points from a collection by their IDs, optionally including stored payloads and vectors.
Query Points Search a collection with the universal Query API: the primary similarity-search operation. Query by a raw embedding or an existing point ID, with payload filtering.
Scroll Points Iterate through all points in a collection that match an optional filter, without similarity scoring, using a page offset.
Upsert Points Insert new points or update existing ones matched by ID, each with a vector and an optional payload of metadata. Waits for changes to apply.

Payload

Clear Payload Remove the entire payload from selected points; the vectors are kept. Select points by IDs or by filter.
Delete Payload Keys Remove specific payload fields from selected points; other payload fields and the vectors are kept.
Overwrite Payload Replace the entire payload of selected points with the provided object; all previously stored payload fields are removed.
Set Payload Set (merge) payload fields on selected points; existing keys not present in the new payload are kept.

Start building with Qdrant

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

Start Building Free Talk to us