# Qdrant

> This skill should be used when the user asks to inspect, operate, or troubleshoot a Qdrant vector database. Triggers include: "list Qdrant collections", "create a collection", "inspect collection config", "upsert points", "query points", "run similarity search", "find nearest neighbors", "check Qdrant health", "embedding storage", or "vector database is down".

- Skill: `dinglebear-ai/qdrant` (Agent Skill, multi-file: 5 files)
- Install (CLI): `npx skillmds@latest add dinglebear-ai/qdrant`
- Raw SKILL.md: https://api.skillmd.com/api/skills/dinglebear-ai/qdrant/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: dinglebear-ai (https://skillmd.com/u/dinglebear-ai)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/dinglebear-ai/qdrant

---


# Qdrant

Vector database for semantic search and embeddings. Use the Qdrant REST API for
direct collection, point, and search operations.

## How to call it

Prefer `scripts/qdrant-api.sh` for common collection and point operations. It
loads `QDRANT_URL` and optional `QDRANT_API_KEY` from the plugin-generated
config first, then the environment or legacy `~/.lab/.env`, adds the API key
header only when present, and exposes `health`,
`collections`, `collection`, `create`, `scroll`, `query`, `upsert`, and
`delete-collection`.

Configure `qdrant_url` and optional sensitive `qdrant_api_key` in Claude plugin
settings or Gemini extension settings. The hook writes
`${XDG_CONFIG_HOME:-~/.config}/lab-qdrant/config.env` with mode `600`.

Read connection settings without printing secrets, and do not add real API keys
to repo examples or committed files.

```bash
read_lab_env() {
  awk -F= -v key="$1" '$1 == key { sub(/^[^=]*=/, ""); print; exit }' ~/.lab/.env 2>/dev/null
}

source "${XDG_CONFIG_HOME:-$HOME/.config}/lab-qdrant/config.env" 2>/dev/null || true
QDRANT_URL=${QDRANT_URL:-$(read_lab_env QDRANT_URL)}
QDRANT_API_KEY=${QDRANT_API_KEY:-$(read_lab_env QDRANT_API_KEY)}
: "${QDRANT_URL:?Set QDRANT_URL in generated config, environment, or ~/.lab/.env}"
AUTH=(); [ -n "$QDRANT_API_KEY" ] && AUTH=(-H "api-key: $QDRANT_API_KEY")
```

`QDRANT_API_KEY` is optional. Include the `api-key` header only when it is set.
Never echo the key or paste it into chat.

## Common operations

| Intent | Request |
|---|---|
| Server health / version | `curl -sS "${AUTH[@]}" "$QDRANT_URL/"` |
| List collections | `curl -sS "${AUTH[@]}" "$QDRANT_URL/collections"` |
| Collection info | `curl -sS "${AUTH[@]}" "$QDRANT_URL/collections/<name>"` |
| Create collection | `curl -sS -X PUT "${AUTH[@]}" -H 'Content-Type: application/json' "$QDRANT_URL/collections/<name>" -d '{"vectors":{"size":<dim>,"distance":"Cosine"}}'` |
| Delete collection (**destructive**) | `curl -sS -X DELETE "${AUTH[@]}" "$QDRANT_URL/collections/<name>"` |
| Upsert points | `curl -sS -X PUT "${AUTH[@]}" -H 'Content-Type: application/json' "$QDRANT_URL/collections/<name>/points?wait=true" -d '{"points":[{"id":1,"vector":[...],"payload":{}}]}'` |
| Query nearest points | `curl -sS -X POST "${AUTH[@]}" -H 'Content-Type: application/json' "$QDRANT_URL/collections/<name>/points/query" -d '{"query":[...],"limit":10,"with_payload":true}'` |
| Scroll points | `curl -sS -X POST "${AUTH[@]}" -H 'Content-Type: application/json' "$QDRANT_URL/collections/<name>/points/scroll" -d '{"limit":10,"with_payload":true,"with_vector":false}'` |

Full REST reference: <https://api.qdrant.tech/>

## Destructive actions

Deleting a collection (`DELETE /collections/<name>`) removes all of its data and
is irreversible. Confirm with the user before running it. Also confirm before
large point deletions, shard changes, snapshot restores, or collection
recreation.

## Configuration

`QDRANT_URL` is required. `QDRANT_API_KEY` is optional. Prefer Claude/Gemini
settings; use `~/.lab/.env` only as a local runtime fallback, not as committed
configuration. Verify connectivity:

```bash
curl -sS "${AUTH[@]}" "$QDRANT_URL/" -w '\nHTTP %{http_code}\n'
```

If the local Labby gateway exposes a Qdrant tool, use that managed tool instead
of raw curl after confirming its live schema. Raw REST remains the fallback for
direct instance work.

## When NOT to use this skill

- The user is asking about a different homelab service — load that service's skill instead.
- The user wants to *generate* embeddings, not store or search them - use the
  `tei` skill.