Qdrant(@qdrant_engine)

A reminder that we’ve got an amazing series of events lined up for you all next week We’ve got thre...

Qdrant (@qdrant_engine) — Fri, 11 Sep 2026 15:00:00 GMT

A reminder that we’ve got an amazing series of events lined up for you all next week

We’ve got three community events happening across September 16-17:

→ Sep 16: Discord Office Hours, Virtual
An open discussion with the Qdrant team. No fixed agenda, bring your questions, share what you’re building, or just drop in and listen.

→ Sep 16: Bring Your Own Demo Night, In person in Berlin
Six community members, five minutes each, showing what they’ve built around vectors, embeddings, and agents. Pizza, beer, and networking with cognee.

→ Sep 17: Vector Space Stream, Virtual
4+ hours of technical talks covering token-native storage, vector compression, robotics, new embedding geometries, hybrid search, and extreme-scale evaluation.

All three are free to join.

Register here: qdrant.tech/events/

💬1🔄3❤️6👀306📊3

⚡ Powered by xgo.ing

We’ve just launched the Qdrant Beginner Course, built for people who want to understand vector searc...

Qdrant (@qdrant_engine) — Wed, 09 Sep 2026 16:34:28 GMT

We’ve just launched the Qdrant Beginner Course, built for people who want to understand vector search from the ground up, with 𝐧𝐨 𝐩𝐫𝐢𝐨𝐫 𝐞𝐱𝐩𝐞𝐫𝐢𝐞𝐧𝐜𝐞 𝐫𝐞𝐪𝐮𝐢𝐫𝐞𝐝.

The course starts with the basics: why traditional search struggles, how embeddings represent meaning, and how similarity works. From there, you’ll build your first Qdrant collection, learn about vectors, payloads and HNSW, and understand the differences between dense, sparse, and hybrid search.

You’ll also learn how to design a complete vector search system, including filtering, RAG, and deployment considerations, before putting everything together in a hands-on multimodal supplier risk intelligence capstone project.

The core course takes under 2 hours, with the capstone taking around 3 more hours. It’s self-paced, hands-on, and you can earn a 𝐬𝐡𝐚𝐫𝐞𝐚𝐛𝐥𝐞 𝐜𝐞𝐫𝐭𝐢𝐟�qdrant.tech/course/beginne…lete it.

Start here: https://t.co/GEjzxAgMbB

💬2🔄3❤️13👀698📊5

⚡ Powered by xgo.ing

Your search can be working perfectly and still be badly tuned. Vector search has a lot of tuning kn...

Qdrant (@qdrant_engine) — Tue, 08 Sep 2026 08:51:15 GMT

Your search can be working perfectly and still be badly tuned.

Vector search has a lot of tuning knobs: `hnsw_ef`, candidate depth, RRF `k`, quantization, reranking, and more.

The problem is, when search quality isn't where you want it to be, which knob should you change first?

We tested this across five public datasets and found that the answer depends on where the problem actually is.

For example, increasing candidate depth from 10 → 500 improved the best achievable score by up to 0.28. But the final score improved by at most 0.01 because the relevant documents were already being retrieved, they just weren't ranking high enough.

So instead of changing settings until the score improves, first identify what's failing, then tune the knob connected to that problem.

That's what we explore in our latest article: how to systematically tune retrieval without guessing.

qdrant.tech/blog/tuning-re…

💬6🔄1❤️16👀1323📊10

⚡ Powered by xgo.ing

Next week is packed at Qdrant. We’ve got 3 community events happening across Sept 16-17, covering e...

Qdrant (@qdrant_engine) — Mon, 07 Sep 2026 11:56:17 GMT

Next week is packed at Qdrant.

We’ve got 3 community events happening across Sept 16-17, covering everything from open discussions and community demos to the latest research in vector search.

Sept 16:
• Discord Office Hours: an open conversation with the Qdrant team. Bring your questions or talk about what you’re building.
• Bring Your Own Demo Night in Berlin with @cognee_: 6 community builders, 5 minutes each to demo what they’ve built with vectors, embeddings, or agents. Plus pizza, drinks & networking.

Sept 17:
• Vector Space Stream: 4+ hours of technical talks on vector compression, Qdrant Edge, embedding geometries, hybrid search, and extreme-scale vector evaluation.

No marketing pitches. Just technical talks, research, experiments & builders.

All 3 are free → qdrant.tech/events

💬3🔄4❤️18👀1051📊7

⚡ Powered by xgo.ing

We just released Qdrant 1.19.1, a patch release with some serious performance improvements under the...

Qdrant (@qdrant_engine) — Fri, 04 Sep 2026 16:47:07 GMT

We just released Qdrant 1.19.1, a patch release with some serious performance improvements under the hood.

Quantized HNSW search is now ~1.5–2.5× faster, and payload-heavy shard transfers are 1.5× faster.

Plus improvements to TurboQuant SIMD scoring, batched HNSW search, deletion checks, and more.

Release here: github.com/qdrant/qdrant/…

💬1🔄4❤️23👀1108📊5

⚡ Powered by xgo.ing

@huggingface 7/ Compute from @Vultr. Distribution with @skypilot_org. Corpus and hosting from @huggi...

Qdrant (@qdrant_engine) — Tue, 01 Sep 2026 13:09:37 GMT

@huggingface 7/ Compute from @Vultr. Distribution with @skypilot_org. Corpus and hosting from @huggingface.

The era of 1M-vector datasets and synthetic approximations is over. Real data, real ground truth, open tooling. All links in the first post.

💬0🔄0❤️5👀192📊1

⚡ Powered by xgo.ing

6/ Supernova, the open-source engine that we used, is fully open source and free. Four phases: embe...

Qdrant (@qdrant_engine) — Tue, 01 Sep 2026 13:09:07 GMT

6/ Supernova, the open-source engine that we used, is fully open source and free.

Four phases: embedding generation, GPU-native brute-force ground truth, cluster loading, evaluation. YAML-configured, scales across @awscloud, @googlecloud, @Azure, @kubernetesio, and Slurm via @skypilot_org.

💬0🔄0❤️2👀82

⚡ Powered by xgo.ing

@huggingface 5/ Two companion datasets ship alongside it: PubMed-Multi-Vector: dense, sparse, and C...

Qdrant (@qdrant_engine) — Tue, 01 Sep 2026 13:08:19 GMT

@huggingface 5/ Two companion datasets ship alongside it:

PubMed-Multi-Vector: dense, sparse, and ColBERT-style multi-vector over one corpus. 8.37B multi-vector tokens.

Coyo-Vector-Embeddings: multimodal image-caption embeddings at 2,048 dimensions.

💬0🔄0❤️4👀128📊1

⚡ Powered by xgo.ing

@huggingface 4/ The 120,000 queries cover dense, sparse, and filtered search, so you can measure wha...

Qdrant (@qdrant_engine) — Tue, 01 Sep 2026 13:08:10 GMT

@huggingface 4/ The 120,000 queries cover dense, sparse, and filtered search, so you can measure what production runs: deep retrieval and filters against exact answers, not 90% recall@10 on a synthetic corpus.

💬0🔄0❤️2👀82

⚡ Powered by xgo.ing

@huggingface 3/ Qdrant-FineWeb-10B, the largest open-source vector search benchmark dataset, is live...

Qdrant (@qdrant_engine) — Tue, 01 Sep 2026 13:08:04 GMT

@huggingface 3/ Qdrant-FineWeb-10B, the largest open-source vector search benchmark dataset, is live on Hugging Face:

10B FineWeb documents
Dense + sparse vectors from gte-multilingual-base
24.47 TB of vectors, 28.66 TB of text and metadata
Exact top-1,000 ground truth for 120,000 queries

💬0🔄0❤️2👀126

⚡ Powered by xgo.ing

@huggingface 2/ Why was this unsolved? Exact ground truth at billion scale means comparing every que...

Qdrant (@qdrant_engine) — Tue, 01 Sep 2026 13:07:47 GMT

@huggingface 2/ Why was this unsolved? Exact ground truth at billion scale means comparing every query against every vector. At 10B documents, over a quadrillion distance computations.

We like hard problems. With @Vultr, we absorbed that cost so the community doesn't have to.

💬0🔄0❤️4👀242📊1

⚡ Powered by xgo.ing

1/ Production vector search: billions of vectors, thousands of RPS, tight tail latency. Public bench...

Qdrant (@qdrant_engine) — Tue, 01 Sep 2026 13:07:25 GMT

1/ Production vector search: billions of vectors, thousands of RPS, tight tail latency. Public benchmarks don't come close. So we built what was missing and gave it away.

Write-up: qdrant.tech/blog/qdrant-fi…
Dataset on @huggingface: huggingface.co/datasets/Qdran…
Supernova Code: github.com/qdrant-labs/su…

💬9🔄9❤️48👀7235📊21

⚡ Powered by xgo.ing

Vector Space Stream - September 17 - Virtual Register: https://t.co/qlSQ3Tdtlo

Qdrant (@qdrant_engine) — Mon, 31 Aug 2026 17:47:11 GMT

Vector Space Stream - September 17 - Virtual

Register: luma.com/vector-space-s…

💬0🔄0❤️1👀114

⚡ Powered by xgo.ing

(6/7) And there’s a session on hybrid search tuning. The work benchmarks a hybrid pipeline across 5...

Qdrant (@qdrant_engine) — Mon, 31 Aug 2026 17:46:44 GMT

(6/7)
And there’s a session on hybrid search tuning.

The work benchmarks a hybrid pipeline across 5 datasets to understand which tuning parameters actually move retrieval quality.

Not just what each knob does.

Which knobs actually matter on real data.

💬0🔄0❤️0👀61

⚡ Powered by xgo.ing

(5/7) The research goes beyond traditional vector search too. We’ll look at: → Hyperbolic and spher...

Qdrant (@qdrant_engine) — Mon, 31 Aug 2026 17:46:30 GMT

(5/7)
The research goes beyond traditional vector search too.

We’ll look at:
→ Hyperbolic and spherical geometries
→ New approaches to embeddings
→ Asymmetric dual encoders
→ Smaller on-device query encoders
→ Efficient edge retrieval architectures

💬0🔄0❤️0👀23

⚡ Powered by xgo.ing

(4/7) Then there’s robotics. Reachy + Qdrant Edge explores what it takes to give a robot memory tha...

Qdrant (@qdrant_engine) — Mon, 31 Aug 2026 17:46:11 GMT

(4/7)
Then there’s robotics.

Reachy + Qdrant Edge explores what it takes to give a robot memory that runs directly on its own hardware.

Embedded. In-process. No cloud.

The session looks at memory and recognition while staying fully offline.

💬0🔄0❤️0👀30

⚡ Powered by xgo.ing

(3/7) Vector compression is another major topic. The stream covers the research behind TurboQuant a...

Qdrant (@qdrant_engine) — Mon, 31 Aug 2026 17:45:59 GMT

(3/7)
Vector compression is another major topic.

The stream covers the research behind TurboQuant and the turbo4 datatype, including how vector compression can be applied to multivector search while maintaining high recall.

💬0🔄0❤️0👀69

⚡ Powered by xgo.ing

(2/7) One session explores token-native storage, with research showing: 2–3× compression 10–100× fa...

Qdrant (@qdrant_engine) — Mon, 31 Aug 2026 17:45:43 GMT

(2/7)
One session explores token-native storage, with research showing:

2–3× compression
10–100× faster reads
2–20× faster writes

for scaled workloads.

💬0🔄0❤️1👀86

⚡ Powered by xgo.ing

A few hours of deep dives into what @qdrant_engine researchers and engineers are actually working on...

Qdrant (@qdrant_engine) — Mon, 31 Aug 2026 17:45:17 GMT

A few hours of deep dives into what @qdrant_engine researchers and engineers are actually working on.

We are hosting Vector Space Stream, that is happening on September 17.

Here’s what’s on the agenda. 🧵

💬7🔄4❤️8👀1136📊9

⚡ Powered by xgo.ing

What if warehouse robots could search product catalogs without the network? Rahul Sharma shows how ...

Qdrant (@qdrant_engine) — Tue, 25 Aug 2026 08:07:23 GMT

What if warehouse robots could search product catalogs without the network?

Rahul Sharma shows how Qdrant Edge + MobileCLIP2 enables offline visual search, with multivectors helping recognize products across different views.

Read more: levelup.gitconnected.com/solving-the-mi…

💬2🔄5❤️18👀838📊7

⚡ Powered by xgo.ing

Only 2 days to go! We’re coming to New York for an AI debate night where you’ll defend the side you ...

Qdrant (@qdrant_engine) — Mon, 24 Aug 2026 18:00:00 GMT

Only 2 days to go! We’re coming to New York for an AI debate night where you’ll defend the side you didn’t choose.

Join us for Hard Negatives: AI Engineers Debate & Game Night.

Draw a controversial AI take at random, get 3 minutes to prepare, and 10 minutes to convince the room. The audience votes live, and the strongest arguments win prizes.

Time: 6:30 PM – 10:00 PM EDT

Limited spots. 21+. Registration approval required.

Register: luma.com/nyc-meetup-qdr…

Your browser does not support the video tag.

🔗 View on Twitter

💬0🔄1❤️7👀524📊1

⚡ Powered by xgo.ing

How to cut RAG token costs by 67% without adding another reranking service? @cappybaradeploy combin...

Qdrant (@qdrant_engine) — Mon, 24 Aug 2026 14:12:30 GMT

How to cut RAG token costs by 67% without adding another reranking service?

@cappybaradeploy combines Qdrant’s native ColBERT reranking, binary quantization, and sentence-level retrieval to send only the most relevant parts of a document to the LLM.

The benchmark achieved 67.1% fewer input tokens, while keeping reranking inside Qdrant instead of adding an external API.

A practical deep dive into building more efficient RAG pipelines.

Read the full article:
pub.towardsai.net/how-qdrant-red…

💬4🔄1❤️36👀1472📊11

⚡ Powered by xgo.ing

What if your AI Agent didn’t need to answer the same question twice? Semantic caching can recognize...

Qdrant (@qdrant_engine) — Thu, 20 Aug 2026 15:30:00 GMT

What if your AI Agent didn’t need to answer the same question twice?

Semantic caching can recognize when two differently worded questions have the same meaning and reuse an existing response instead of calling the LLM again.

In this practical deep dive, Tina Sharma builds a semantic cache with Qdrant, benchmarks it, tunes similarity thresholds, and compares single-vector vs. multi-vector retrieval.

The results: 57.1% cache hit rate, 55.7% fewer tokens, and ~15ms response time for cache hits.

Read the full article and explore the implementation:

ai.gopubby.com/how-to-use-sem…

💬1🔄2❤️13👀596📊4

⚡ Powered by xgo.ing

Today’s the day! Let’s talk about Qdrant 1.19. Join us for our Qdrant Discord Office Hours as we wa...

Qdrant (@qdrant_engine) — Thu, 20 Aug 2026 12:06:58 GMT

Today’s the day! Let’s talk about Qdrant 1.19.

Join us for our Qdrant Discord Office Hours as we walk through the latest features and updates in 1.19, followed by an open Q&A and community discussion.

Time: 5:00 PM CEST
Join here: discord.gg/PGemY6ugf?even…

Have questions about 1.19? Bring them along, we’d love to hear what you’re working on.

More on Qdrant 1.19: qdrant.tech/blog/qdrant-1.…

See you there!

💬0🔄1❤️10👀531📊2

⚡ Powered by xgo.ing

Less than a week to go until Hard Negatives in San Francisco. Join Qdrant and @neo4j for an evening...

Qdrant (@qdrant_engine) — Thu, 20 Aug 2026 02:30:00 GMT

Less than a week to go until Hard Negatives in San Francisco.

Join Qdrant and @neo4j for an evening of AI debates, games, food, drinks, and prizes.

You’ll be paired into teams and randomly assigned a controversial AI topic to defend. You get a few minutes to prepare, then it’s time to convince the room. The audience decides who wins.

Prizes:
- First prize: Nintendo Switch
- Second prize: Surprise LEGO set
- Third prize: Qdrant hoodie

August 25
 6:30 PM - 8:30 PM PDT
Register: luma.com/nyc-meetup-qdr…

💬1🔄2❤️4👀820📊2

⚡ Powered by xgo.ing

We’re coming to Tokyo for a night of fast building, AI agents, and working demos. Join us for Fast ...

Qdrant (@qdrant_engine) — Wed, 19 Aug 2026 13:30:00 GMT

We’re coming to Tokyo for a night of fast building, AI agents, and working demos.

Join us for Fast Hacks: Hack Night with Qdrant, @OpenAI Codex, @HackerSquad and @neo4j on August 27.

You’ll have two hours to build with your AI agents, then showcase your project in front of the community. Come with an idea, find a team on the spot, or just come to watch, learn, and meet other builders.

Our Developer Relations Engineer Ewa Szyszka will be joining the event as one of the judges.

Stay tuned for more Qdrant events coming to Tokyo!

Register: hackersquad.io/events/openai-…

💬2🔄1❤️12👀613📊4

⚡ Powered by xgo.ing

ACORN fixes filtered search at query time. Qdrant's filterable HNSW fixes it in the index. Filters ...

Qdrant (@qdrant_engine) — Wed, 19 Aug 2026 08:54:24 GMT

ACORN fixes filtered search at query time. Qdrant's filterable HNSW fixes it in the index.

Filters can break an HNSW graph into disconnected islands, leaving search stranded and recall behind. ACORN works around that at query time by stepping through filtered-out neighbors.

Filterable HNSW takes a different approach: it adds edges between points that share an indexed payload value, so the filtered graph remains connected.

On a 1% filter over 1M vectors, filterable HNSW reached 99.8% recall at 1.0ms. ACORN reached 67.7% at 4.7ms.

ACORN still earns its place for gaps like broad values and AND filters. But it's most effective when it has a graph built for filters underneath it.

Full benchmark: qdrant.tech/articles/filte…

💬0🔄5❤️19👀1172📊5

⚡ Powered by xgo.ing

Qdrant + Minima: agentic RAG on one RTX PRO 6000 Blackwell GPU. Hybrid search and late-interaction ...

Qdrant (@qdrant_engine) — Fri, 14 Aug 2026 19:22:50 GMT

Qdrant + Minima: agentic RAG on one RTX PRO 6000 Blackwell GPU.

Hybrid search and late-interaction reranking made the first retrieval enough 87% of the time, up from 72%.

Median latency 21.3s to 7.7s.

2.92x more successful tasks per GPU-hour.

qdrant.tech/blog/case-stud…

💬0🔄5❤️19👀936📊6

⚡ Powered by xgo.ing

We’re coming to New York for an AI debate night where you’ll defend the side you didn’t choose. Joi...

Qdrant (@qdrant_engine) — Fri, 14 Aug 2026 12:37:32 GMT

We’re coming to New York for an AI debate night where you’ll defend the side you didn’t choose.

Join us for Hard Negatives: AI Engineers Debate & Game Night.

You’ll draw a controversial AI take at random, get 3 minutes to prepare, and 10 minutes to convince the room. The audience votes live, and the strongest arguments win prizes.

Prizes, food, drinks, pool, darts, foosball, ping pong, and shuffleboard included.

Date: August 26
Time: 6:30 PM - 10:00 PM EDT
Venue: Sugar Mouse NYC

Limited spots. 21+. Registration approval required.
Register: luma.com/nyc-meetup-qdr…

💬4🔄1❤️4👀489📊5

⚡ Powered by xgo.ing

How @Bayer built an enterprise search engine for 116,000 employees on Qdrant Hybrid Cloud:1 35M poi...

Qdrant (@qdrant_engine) — Thu, 13 Aug 2026 18:17:10 GMT

How @Bayer built an enterprise search engine for 116,000 employees on Qdrant Hybrid Cloud:1

35M points on 4 nodes. Hybrid search for deep agents. Semantic caching to control agent cost. A measured 20% efficiency gain.

Three years in production, from prototype to platform: qdrant.tech/blog/case-stud…

💬2🔄1❤️15👀545📊5

⚡ Powered by xgo.ing

We’re coming to San Francisco for a night of AI debates, games, and prizes. Join Qdrant and @neo4j ...

Qdrant (@qdrant_engine) — Thu, 13 Aug 2026 15:00:02 GMT

We’re coming to San Francisco for a night of AI debates, games, and prizes.

Join Qdrant and @neo4j for Hard Negatives Meetup: Engineers Debate Night.

You’ll be paired into teams and randomly assigned a controversial AI topic to defend. You’ll get a few minutes to prepare, then it’s time to convince the room. The audience decides who wins.

Prizes:
- First prize: Nintendo Switch
- Second prize: Surprise LEGO set
- Third prize: Qdrant hoodie

Food, drinks, games, and networking with AI engineers, researchers, and builders.

Date: August 25
Time: 6:30 PM - 8:30 PM PDT

Register: luma.com/nyc-meetup-qdr…

Your browser does not support the video tag.

🔗 View on Twitter

💬0🔄3❤️4👀746📊2

⚡ Powered by xgo.ing

Let’s Talk about Qdrant 1.19! It’s time to discuss what’s new. Join us for our next Qdrant Discord...

Qdrant (@qdrant_engine) — Thu, 13 Aug 2026 06:00:01 GMT

Let’s Talk about Qdrant 1.19!

It’s time to discuss what’s new.

Join us for our next Qdrant Discord Office Hours as we walk through the latest features and updates, followed by an open Q&A and community discussion.

Date: August 20
Time: 5:00 PM CEST
Link: discord.gg/PGemY6ugf?even…

Bring your questions, projects, and ideas. We’d love to hear what you’re working on!

More details about 1.19 update: qdrant.tech/blog/qdrant-1.…

See you there!

💬1🔄2❤️12👀1184📊5

⚡ Powered by xgo.ing

2 Days Until We Give Slack a Memory Join Qdrant and @cognee_ in Berlin for a hands-on hack night wh...

Qdrant (@qdrant_engine) — Wed, 12 Aug 2026 18:37:49 GMT

2 Days Until We Give Slack a Memory

Join Qdrant and @cognee_ in Berlin for a hands-on hack night where you’ll build a memory layer that helps AI agents reason over your Slack history.

-  €1,200 prize pool
-  Swag, datasets, API keys + support from both teams
- Bring your laptop and build with us!

Date: August 14
Location: Paul-Lincke-Ufer 39-40, Berlin

Register now: luma.com/cognee-m078

💬1🔄2❤️6👀482📊3

⚡ Powered by xgo.ing

Keyword indexes in Qdrant are built for exact matching. They’re ideal for values like URLs, file pat...

Qdrant (@qdrant_engine) — Tue, 11 Aug 2026 14:21:04 GMT

Keyword indexes in Qdrant are built for exact matching. They’re ideal for values like URLs, file paths, SKUs, and unique identifiers.

The challenge? If you wanted to find values that start with something, for example, all URLs beginning with “https://qdrant”., you had two options:

- Scan every value in the payload, or
- Switch to a text index, which tokenizes the value and no longer supports exact matching.

Qdrant 1.19 solves this.

You can now enable prefix matching on keyword indexes with "prefix": true in the index configuration. After that, prefix filters are served directly from the index, giving you fast prefix lookups while keeping exact matching.

Full release here: qdrant.tech/blog/qdrant-1.…

💬4🔄4❤️24👀1086📊9

⚡ Powered by xgo.ing

BM25 scoring relies on IDF (Inverse Document Frequency). In simple terms, the rarer a word is in you...

Qdrant (@qdrant_engine) — Mon, 10 Aug 2026 15:29:33 GMT

BM25 scoring relies on IDF (Inverse Document Frequency). In simple terms, the rarer a word is in your data, the more important it becomes during search.

In a multi-tenant collection, IDF was previously calculated across all tenants combined. That meant if Tenant A and Tenant B used very different vocabularies, their statistics were mixed together. As a result, a word that was rare for one tenant might not be treated as rare, leading to less accurate BM25 scores.

Qdrant 1.19 fixes this.

You can now calculate IDF statistics per tenant instead of across the entire collection. Each tenant’s BM25 scores are now based only on their own data, resulting in more accurate lexical search without requiring separate collections.

Full release here: qdrant.tech/blog/qdrant-1.…

💬0🔄6❤️31👀1563📊7

⚡ Powered by xgo.ing

Qdrant 1.18 introduced TurboQuant, a quantization technique that speeds up vector search by keeping ...

Qdrant (@qdrant_engine) — Fri, 07 Aug 2026 06:51:58 GMT

Qdrant 1.18 introduced TurboQuant, a quantization technique that speeds up vector search by keeping a 4-bit compressed copy of your vectors in memory while storing the original float32 vectors on disk for rescoring.

Great recall, but two copies of every vector.

We just released Qdrant 1.19, which introduces Turbo4, a new datatype that stores only the 4-bit representation, removing the original float32 copy entirely.

That reduces storage from 36 bits per coordinate to just 4 bits, a 9× storage reduction. With less data to read and write, search throughput also improves.

The tradeoff is simple: without the original vectors, rescoring isn’t possible. If storage efficiency is your priority, Turbo4 is the right choice. If you need the highest possible recall, TurboQuant with full-precision storage is still the better option.

The same storage savings apply to multi-vector collections, making ColBERT-style late interaction search significantly more space-efficient.

Full release here: qdrant.tech/blog/qdrant-1.…

💬2🔄15❤️72👀3902📊19

⚡ Powered by xgo.ing

Can your Slack remember what your team already knows? Join Qdrant and @cognee_ in Berlin for an eve...

Qdrant (@qdrant_engine) — Thu, 06 Aug 2026 11:00:00 GMT

Can your Slack remember what your team already knows?

Join Qdrant and @cognee_ in Berlin for an evening of building AI memory that goes beyond context windows.

At this hands-on hack night, you’ll build a memory layer that turns your Slack workspace into something agents can actually reason over, using cognee for structured memory and Qdrant for fast semantic retrieval.

- €1,200 prize pool:
- Plus, you’ll get swag, API keys, sample datasets, and support from the cognee and Qdrant teams throughout the event.

Date: August 14
Location: Paul-Lincke-Ufer 39-40/hof 4, Berlin

Register now: luma.com/cognee-m078

💬1🔄1❤️7👀554📊2

⚡ Powered by xgo.ing

(6/6) One upgrade note: the legacy /search, /recommend, and /discover endpoints are gone. Migrate to...

Qdrant (@qdrant_engine) — Wed, 05 Aug 2026 22:32:55 GMT

(6/6) One upgrade note: the legacy /search, /recommend, and /discover endpoints are gone. Migrate to /query before you upgrade.

Full release blog here: qdrant.tech/blog/qdrant-1.…

💬0🔄0❤️2👀175

⚡ Powered by xgo.ing

(5/6) Filtering: Prefix matching on keyword indexes means no more full payload scans, and no more tr...

Qdrant (@qdrant_engine) — Wed, 05 Aug 2026 22:32:25 GMT

(5/6) Filtering: Prefix matching on keyword indexes means no more full payload scans, and no more trading exact matching for a text index.

💬0🔄0❤️2👀216

⚡ Powered by xgo.ing

(4/6) Per-Tenant IDF Statistics: You can now narrow the corpus to a single tenant, so term rarity re...

Qdrant (@qdrant_engine) — Wed, 05 Aug 2026 22:32:15 GMT

(4/6) Per-Tenant IDF Statistics: You can now narrow the corpus to a single tenant, so term rarity reflects that tenant's data instead of the whole dataset.

💬0🔄0❤️2👀63

⚡ Powered by xgo.ing

(3/6) Memory Tiers: One memory parameter now replaces on_disk, always_ram, and on_disk_payload, and ...

Qdrant (@qdrant_engine) — Wed, 05 Aug 2026 22:32:00 GMT

(3/6) Memory Tiers: One memory parameter now replaces on_disk, always_ram, and on_disk_payload, and it works the same way on every component of a collection.

💬0🔄0❤️4👀40📊1

⚡ Powered by xgo.ing

(2/6) TurboQuant Datatype: Version 1.18 shipped TurboQuant as a quantization layer. 1.19 enables the...

Qdrant (@qdrant_engine) — Wed, 05 Aug 2026 22:31:45 GMT

(2/6) TurboQuant Datatype: Version 1.18 shipped TurboQuant as a quantization layer. 1.19 enables the new Turbo4 datatype, which keeps only the 4-bit representation. That takes 36 bits per coordinate (the float32 original plus the 4-bit copy) down to four, a ninefold storage reduction, and fewer reads and writes per operation lifts throughput

💬0🔄0❤️4👀123📊1

⚡ Powered by xgo.ing

(1/6) Qdrant 1.19 is now here! Below are the main updates 🧵 Full release blog: https://t.co/v52PS1...

Qdrant (@qdrant_engine) — Wed, 05 Aug 2026 22:31:28 GMT

(1/6) Qdrant 1.19 is now here! Below are the main updates 🧵

Full release blog:qdrant.tech/blog/qdrant-1.…Z

💬5🔄6❤️19👀935📊9

⚡ Powered by xgo.ing

Why does your RAG agent get worse over time? Join @DylanCouzon and Rishabh from FutureAGI tomorrow ...

Qdrant (@qdrant_engine) — Wed, 05 Aug 2026 16:22:11 GMT

Why does your RAG agent get worse over time?

Join @DylanCouzon and Rishabh from FutureAGI tomorrow as they build, debug, and optimize a RAG agent live, covering embedding migrations, deduplication, reranking, and evaluation to measure what actually improves retrieval quality.

Time: 6:30 PM CEST | 9:30 AM PDT | 10:00 PM IST

Register now: luma.com/future-pq6q

💬2🔄2❤️6👀528📊4

⚡ Powered by xgo.ing

You know you are hitting the API and making round trips again and again when a user asks the same qu...

Qdrant (@qdrant_engine) — Tue, 04 Aug 2026 10:54:57 GMT

You know you are hitting the API and making round trips again and again when a user asks the same question, like "what's the refund policy?" or "what are the conditions of a refund?"

You might not have a cache layer built in, or if you do it's probably keyword-based, which only works on exact matches. In that case semantic cache can be useful.

What semantic cache does is store previously asked questions along with their answers. When a new query comes in, it checks for semantic similarity against cached questions. If the similarity score is above a threshold, it returns the cached answer without going through retrieval and generation again. Same answer, no unnecessary API calls or retrieval round trips.

You can learn more about how semantic cache works in practice in our guide: qdrant.tech/articles/seman…

💬1🔄4❤️10👀814📊4

⚡ Powered by xgo.ing

Search. Hybrid retrieval. Faceted filters. Recommendations. Personalization. Visual similarity. Merc...

Qdrant (@qdrant_engine) — Mon, 03 Aug 2026 07:48:55 GMT

Search. Hybrid retrieval. Faceted filters. Recommendations. Personalization. Visual similarity. Merchandising.

All over 5.8M real products.

Performance:
- 40 ms for search with filters and personalization
- 10 ms for similar products
- <1 ms for product lookups
- 96% Precision@10
- ~1.9 GB RAM

One Query API call. That's the whole stack.

The demo and code are both open source. Shoutout to @DylanCouzon for building it. 👏

Demo:demo-ecommerce-search.vercel.appJ
Code:github.com/qdrant-labs/de…N

💬2🔄1❤️14👀738📊4

⚡ Powered by xgo.ing

One week to go! Join us for “Why does my RAG-agent get worse? A live autopsy” with @DylanCouzon fro...

Qdrant (@qdrant_engine) — Fri, 31 Jul 2026 15:00:02 GMT

One week to go!

Join us for “Why does my RAG-agent get worse? A live autopsy” with @DylanCouzon from Qdrant and Rishabh from FutureAGI.

As your RAG system evolves, quality can quietly degrade. In this live session, we’ll build, debug, and improve a RAG agent while measuring the impact of every change.

We’ll cover:
→ Finding where your agent is losing quality
→ Migrating to a new embedding model with zero downtime
→ Removing duplicate and fragmented chunks
→ Adding late interaction reranking
→ Comparing evaluation scores before and after each optimization

We’re also looking for 5-minute community demos before the webinar. If you’d like to present your project, apply here: docs.google.com/forms/d/10-9oy…

Date: August 6
Time: 6:30 PM CEST | 9:30 AM PDT | 10:00 PM IST

Register now: luma.com/future-pq6q

💬1🔄1❤️9👀505📊3

⚡ Powered by xgo.ing

We’re coming to New York for AI debates where you’ll defend the side you didn’t choose. Join us for...

Qdrant (@qdrant_engine) — Thu, 30 Jul 2026 19:20:27 GMT

We’re coming to New York for AI debates where you’ll defend the side you didn’t choose.

Join us for Hard Negatives: AI Engineers Debate & Game Night, where controversial AI takes are assigned at random, and it’s your job to convince the room.

Debate topics like:
→ “RAG is dead”
→ “AI is already sentient”
→ “Open weights should be banned”

- Win a Nintendo Switch, LEGO sets, Qdrant hoodies & more
- Drinks, appetizers, and arcade games

Meet AI engineers, researchers, and builders from across NYC

Date: August 26 | Venue: Sugar Mouse NYC

Register now: luma.com/nyc-meetup-qdr…

💬1🔄4❤️9👀501📊4

⚡ Powered by xgo.ing

Bulk uploading at scale doesn't have to be expensive. A few things worth knowing if you're ingestin...

Qdrant (@qdrant_engine) — Thu, 30 Jul 2026 14:28:22 GMT

Bulk uploading at scale doesn't have to be expensive.

A few things worth knowing if you're ingesting at scale:

Dense vectors may eat RAM fast. The fix? Store the original vectors on disk and keep only the compressed ones in RAM. TurboQuant can get you 8x smaller with almost no recall loss, or 32x if you're really tight on memory.

Sparse vectors have a different case. It's not the vectors that get large, it's the index. You can store that on disk too and use float16 precision to cut it in half.

Create payload indexes before uploading. Qdrant's HNSW builds filter-aware edges during graph construction. Create them after and the graph has to rebuild from scratch.

Full guide: qdrant.tech/articles/bulk-…

💬0🔄1❤️8👀502📊2

⚡ Powered by xgo.ing

Most people treat video like a bag of frames or a transcript with timestamps. But you end up missing...

Qdrant (@qdrant_engine) — Wed, 29 Jul 2026 07:56:23 GMT

Most people treat video like a bag of frames or a transcript with timestamps. But you end up missing the most important parts of it:
- Motion
- Causality
- Temporal progression
- The relationship between what happens and why it matters

At Vector Space Day SF, James Le from @twelve_labs walked through 3 failure modes that come from this:
- Wrong Context
- Wrong Memory
- Wrong Reasoning

And he introduced the stack built to fix it:
→ Marengo: a retrieval model that makes video searchable
→ Pegasus: a video language model that turns retrieved moments into structured answers
→ Jockey: an agentic framework and memory layer for video corpus workflows

With Qdrant handling the storage and retrieval underneath it all.

Full talk is live on our YouTube channel: youtube.com/watch?v=i8xZeK…

💬0🔄2❤️10👀527📊2