Semantic search over Qdrant Cloud collections using the Python client. Covers vector search, filtered search, prefetch+RRF fusion, group API, recommend API, discovery API, batch queries, scroll, and payload indexing. Use when building search features, adding new query types, or working with qdrant-client.
The English summary is being prepared. Showing the original description.
See how to installReview the instructions the agent will receive before you install.
All projects use qdrant-client with 768-dim vectors from nomic-embed-text.
import os
from qdrant_client import QdrantClient, models
client = QdrantClient(
url=os.getenv("QDRANT_URL"),
api_key=os.getenv("QDRANT_API_KEY"),
)
results = client.query_points(
collection_name="DocumentChunk_text",
query=embedding_vector, # list[float], 768-dim
limit=10,
with_payload=True,
)
for point in results.points:
print(point.id, point.score, point.payload)
results = client.query_points(
collection_name="DocumentChunk_text",
query=embedding_vector,
query_filter=models.Filter(
must=[
models.FieldCondition(
key="vendor",
match=models.MatchValue(value="Acme"),
)
]
),
limit=10,
)
Multi-stage retrieval combining multiple query vectors:
results = client.query_points(
collection_name="DocumentChunk_text",
prefetch=[
models.Prefetch(query=vector_a, limit=20),
models.Prefetch(query=vector_b, limit=20),
],
query=models.FusionQuery(fusion=models.Fusion.RRF),
limit=10,
)
Group results by a payload field:
results = client.query_points_groups(
collection_name="DocumentChunk_text",
query=embedding_vector,
group_by="vendor",
limit=5,
group_size=3,
)
Find similar items by point IDs:
results = client.recommend(
collection_name="DocumentChunk_text",
positive=[point_id_1, point_id_2],
negative=[point_id_3],
limit=10,
strategy=models.RecommendStrategy.BEST_SCORE,
)
Guided exploration with context pairs:
results = client.discover(
collection_name="DocumentChunk_text",
target=point_id,
context=[
models.ContextPair(positive=pos_id, negative=neg_id),
],
limit=10,
)
Multiple queries in a single request:
results = client.query_batch_points(
collection_name="DocumentChunk_text",
requests=[
models.QueryRequest(query=vector_1, limit=5),
models.QueryRequest(query=vector_2, limit=5),
],
)
Bulk retrieval without ranking:
records, next_offset = client.scroll(
collection_name="DocumentChunk_text",
limit=100,
with_payload=True,
with_vectors=True,
)
# Paginate with offset=next_offset
Create indexes for fast filtering:
client.create_payload_index(
collection_name="DocumentChunk_text",
field_name="vendor",
field_schema=models.PayloadSchemaType.KEYWORD,
)
# Full-text index
client.create_payload_index(
collection_name="DocumentChunk_text",
field_name="text",
field_schema=models.TextIndexParams(
type="text",
tokenizer=models.TokenizerType.WORD,
min_token_len=2,
max_token_len=15,
lowercase=True,
),
)
No local model needed — Qdrant embeds on the server:
from qdrant_client.http.models import Document
client = QdrantClient(url=QDRANT_URL, api_key=QDRANT_API_KEY, cloud_inference=True)
results = client.query_points(
collection_name="my_collection",
query=Document(text="search query", model="Qdrant/Qwen/Qwen3-Embedding-0.6B"),
limit=10,
)
Warning: Cloud inference uses a different embedding model. Existing collections use 768-dim nomic-embed-text vectors — you cannot mix models in the same collection without re-embedding all data.
No reviews yet. Tell others what you thought.
No login needed. Share how it worked for you, ask a question, or note where it did not run.
No comments yet. Be the first to write one.
Use at a cognee hackathon. Watches the participant's experience with cognee throughout the session — errors, slowness, confusion, workarounds, and signs of frustration in how they work with you — and keeps one private feedback file they send to the cognee team at the end.
The English summary is being prepared. Showing the original description.
Use at the Mem Off hackathon. Keeps a running, evidence-backed report on how the participant's memory system performed for their coding agent during the day — what the harness measured, what cognee-report's scorecard says, what was recalled and whether it helped — in ./memoff-report.md and ./memoff-report.json. The report is submitted together with the action results; it never contains quotes, keys, data or names.
The English summary is being prepared. Showing the original description.
Use when reviewing, fixing, or verifying a pull request. Focus on correctness regressions, missing tests, permission mistakes, unsafe behavior, and minimal fixes.
The English summary is being prepared. Showing the original description.
Build spend analytics and dashboards from Qdrant vector data. Covers bulk scroll extraction, aggregation patterns, vendor grouping, and Chart.js visualization. Use when building analytics features, aggregating procurement data, or creating dashboards.
The English summary is being prepared. Showing the original description.
Detect anomalies in procurement data using Qdrant vector operations. Covers centroid-based outlier detection, near-duplicate finding via batch recommend, amount z-score analysis, and vendor variance scoring. Use when building anomaly detection, fraud detection, or data quality features.
The English summary is being prepared. Showing the original description.