Skip to main content
Collections are tables, points are rows, and the point id is the primary key. The editor takes Qdrant REST requests, not SQL, so a vector search such as POST /collections/docs/points/query runs from the editor and its points and scores land in the grid. The Qdrant driver is not in the app. Picking Qdrant in the Choose a Database sheet offers the download before the form opens, and opening a saved Qdrant connection installs it without asking. See Plugins.

​
Quick setup

Click New Connection…, select Qdrant, enter host and port, paste an API Key if the node needs one, then click Save & Connect. There is no Database field. One connection reaches one Qdrant instance, and its collections are the objects. For a node on a private server, open the Network tab and set Connect via to SSH Tunnel, then set Host to the address Qdrant listens on as the SSH server sees it, usually localhost. See SSH Tunneling.

​
Connection settings

Qdrant Cloud requires the API key and HTTPS. Set SSL Mode to Verify Identity with no CA certificate: the certificate is checked against the system trust store.

​
Browsing collections

The sidebar lists collections from GET /collections. Every grid has id first and the vectors last. id is the primary key and read-only. A collection with one unnamed vector shows a vector column. A collection with named or sparse vectors shows one column per name, such as vector.image. A dense vector is a JSON number array, a multivector an array of them, and a sparse vector its indices and values. Payload columns are the top-level indexed fields, then the fields found in the first 100 points. Arrays and objects render as JSON in the cell. An index on a nested path such as metadata.source or tags[] is listed in the Indexes section of the Structure tab, and its value shows inside the metadata or tags column. A field that appears only in later points shows up in a query result but not as a grid column. Pages load through POST /points/query in id order, with the offset applied by Qdrant, so a page always shows the rows at that position now. A node older than 1.10 has no Query API, so its pages walk the point ids from the start.

​
Sorting

Sort by one payload field that has an integer, float or datetime payload index. Points with a value in that field come first, in order. Points without one follow, by id. Every point stays reachable, and the row count still matches the collection. Sorting by id ascending is the default order. These sorts are refused with a message:
  • id descending, a vector, or more than one column.
  • A field that holds a list in any point, since Qdrant would list that point once per value.

​
Filtering

A payload index does not limit what a field holds, so a value matches every type it can be: 42 finds the text "42", the number 42, and a list holding either. The index still speeds up the part of the filter of its own type. Not equals, not in list and not contains also match points the key is missing from. On id, equals and in list become has_id, and contains finds a whole id only. Starts with, ends with and REGEX have no Qdrant equivalent, and a filter Qdrant cannot run reports why instead of returning the whole collection. The raw filter row takes one Qdrant condition as JSON, such as {"key": "city", "match": {"value": "Berlin"}}. Filtered row counts come from POST /points/count with exact set.

​
Editing points

Grid edits are REST requests keyed by id, shown in the review sheet before they run. A cell is written back as the type its column holds: a keyword field stays a string even when it reads 42, and a bool field takes true, false, 1 or 0. A column whose points hold different types, such as text in one and a number in another, has the type json: its text shows in quotes, and a cell saves as the JSON it spells, so "42" stays text and 42 a number. A value that does not fit, such as 4.5 in an integer column or a vector that is not a number array, stops the save and names the column. Truncate deletes every point with an empty filter and keeps the collection. Drop removes the collection.

​
REST editor

Type a request line, then a JSON body on the same line, on the lines under it, or starting on the line and going on under it:
A response that lists points, from query, scroll, search or a retrieve by id, renders as a grid with id, then score when there is one, the payload fields, and any vectors returned. Anything else is shown as formatted JSON. Read-only mode allows GET, POST /collections/{name}/facet, and the read endpoints under /points: query, scroll, search, count, recommend, discover, their batch and group forms, and a retrieve by id. A DELETE, a path ending in delete or clear, and points/batch count as destructive. A snapshot request is flagged as reaching the server’s disk.

​
SSL/TLS

New connections start on Disabled, plain HTTP. Every other mode goes over HTTPS. For a node that enforces mutual TLS, choose a Client Certificate and its Client Key. See SSL/TLS.

​
Limitations

  • SQL is not available. Use the REST editor.
  • gRPC is not used.
  • Payload indexes and collection settings are not edited from Structure. Send the REST request, such as PUT /collections/{name}/index.
  • A field that holds a list in any point cannot be sorted.
  • Adding rows from the grid needs Qdrant 1.15 or later. On an older node, upsert from the REST editor.
  • The plugin is registry-only. It is not on iPhone or iPad.

​
Troubleshooting

​
Must provide an API key

The node was started with an API key and API Key is empty. Paste the key and connect again.

​
Invalid API key or JWT

The key was rejected. The health check reads /collections, which needs the key, so a revoked key also drops a live session.

​
No range index for order_by key

The sorted column has no integer, float or datetime payload index. Create one with PUT /collections/{name}/index, or clear the sort.

​
A row with this key already exists

A new row’s id is already taken. Clear the id cell to get a new UUID, or edit the existing point instead.