Weaviate AI Database(@weaviate_io)

A bigger context window won't fix a broken agentic RAG pipeline. Long contexts can introduce contra...

Weaviate AI Database (@weaviate_io) — Thu, 23 Jul 2026 16:01:42 GMT

A bigger context window won't fix a broken agentic RAG pipeline.

Long contexts can introduce contradictory information, bury relevant tools, and cause later inputs to be ignored. Every retrieved chunk, tool output, instruction, and conversation turn competes for the model's attention.

This is why production agentic RAG is really a 𝗰𝗼𝗻𝘁𝗲𝘅𝘁 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 problem: controlling what information reaches the model, when, and in what form.

In our newest Youtube video, @victorialslocum breaks down the five systems that make up an agentic RAG pipeline:

1️⃣ 𝗤𝘂𝗲𝗿𝘆 𝗔𝘂𝗴𝗺𝗲𝗻𝘁𝗮𝘁𝗶𝗼𝗻 - translating vague, misspelled, or multipart human input into something retrieval and downstream tools can actually use

2️⃣ 𝗥𝗲𝘁𝗿𝗶𝗲𝘃𝗮𝗹 - choosing a chunking strategy that balances precision with enough surrounding context. Small chunks can lose meaning; large chunks consume context and produce noisier embeddings.

3️⃣ 𝗠𝗲𝗺𝗼𝗿𝘆 - separating active short-term context from externally stored long-term memory. Dumping the full conversation history into every request just lets stale details compete with current information.

4️⃣ 𝗧𝗼𝗼𝗹𝘀 - giving the model clear descriptions, schemas, and access to the right tools at the right stage. Many tool failures are actually context engineering failures in disguise.

5️⃣ 𝗔𝗴𝗲𝗻𝘁𝘀 - the decision and orchestration layer connecting everything above. Agents can reformulate queries, select tools, evaluate results, and change strategy, but they also compound every upstream context failure.

A better prompt can't repair irrelevant retrieval, contryoutu.be/sUoNoaatRvU?si…ool the model can't find.

The video: https://t.co/M9I9WKbOjL

💬1🔄0❤️5👀217📊2

⚡ Powered by xgo.ing

𝗙𝗶𝗿𝗲-𝗮𝗻𝗱-𝗳𝗼𝗿𝗴𝗲𝘁 𝗺𝗲𝗺𝗼𝗿𝘆 𝗳𝗼𝗿 𝗺𝘂𝗹𝘁𝗶-𝗮𝗴𝗲𝗻𝘁 𝗽𝗶𝗽𝗲𝗹𝗶𝗻𝗲𝘀? No block...

Weaviate AI Database (@weaviate_io) — Wed, 22 Jul 2026 15:01:31 GMT

𝗙𝗶𝗿𝗲-𝗮𝗻𝗱-𝗳𝗼𝗿𝗴𝗲𝘁 𝗺𝗲𝗺𝗼𝗿𝘆 𝗳𝗼𝗿 𝗺𝘂𝗹𝘁𝗶-𝗮𝗴𝗲𝗻𝘁 𝗽𝗶𝗽𝗲𝗹𝗶𝗻𝗲𝘀?

No blocking. No duplicates. Just continuous learning that works.

When you add data to Engram, pipelines run in the background to:
1️⃣ 𝗘𝘅𝘁𝗿𝗮𝗰𝘁 relevant memories matching your configured topics
2️⃣ 𝗧𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺 by retrieving existing memories and reconciling them (deduplication, handling preference changes, updating facts)
3️⃣ 𝗖𝗼𝗺𝗺𝗶𝘁 clean, structured memories to Weaviate

But how do you deal with a mutli-agent pipeline where 𝘢𝘭𝘭 𝘵𝘩𝘦 𝘢𝘨𝘦𝘯𝘵𝘴 are adding context to the memory system?

Take a multi-agent RAG system where a main agent handles conversations and a search agent uses specialized tools. A user asks about comedy movies, but the search agent does a text search on "comedy" instead of filtering by genre.

The agent then sends back feedback: "Comedy is a genre, you should filter on the 'genres' property."

The information needed to learn is spread across multiple agents and context windows:
• Task goal: "User asked for comedy movies" (main agent)
• Actions taken: "Called search with near-text query 'comedy'" (search agent)
• Feedback: "Should filter on genres property" (main agent)

Engram's pipeline:

1. Extracts each piece individually using different 𝘁𝗼𝗽𝗶𝗰𝘀 (task_goal, actions_taken, feedback)
2. 𝗕𝘂𝗳𝗳𝗲𝗿𝘀 collect these memories until all pieces arrive
3. 𝗧𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺 step combines them into a single actionable memory:

"𝘞𝘩𝘦𝘯 𝘢𝘴𝘬𝘦𝘥 𝘵𝘰 𝘧𝘪𝘯𝘥 𝘮𝘰𝘷𝘪𝘦𝘴 𝘰𝘧 𝘢 𝘱𝘢𝘳𝘵𝘪𝘤𝘶𝘭𝘢𝘳 𝘨𝘦𝘯𝘳𝘦 (𝘦.𝘨., 𝘤𝘰𝘮𝘦𝘥𝘺), 𝘺𝘰𝘶 𝘴𝘩𝘰𝘶𝘭𝘥 𝘧𝘪𝘭𝘵𝘦𝘳 𝘰𝘯 𝘵𝘩𝘦 𝘨𝘦𝘯𝘳𝘦𝘴 𝘱𝘳𝘰𝘱𝘦𝘳𝘵𝘺, 𝘯𝘰𝘵 𝘥𝘰 𝘢 𝘯𝘦𝘢𝘳-𝘵𝘦𝘹𝘵 𝘲𝘶𝘦𝘳𝘺."

This memory is now stored in Weaviate and retrieved when starting similar tasks. The agent has actually learned from experience 🔥

Because pipelines run asynchronously, agents can:
• Learn from feedback over time without blocking
• Combine information spread across multiple conversations
• Build up experience that improves future performance
• Maintain clean, reconciled memories instead of noisy duplicates

With Engram, you can configure memories to be:
• 𝗣𝗿𝗼𝗷𝗲𝗰𝘁-𝘄𝗶𝗱𝗲 𝘀𝗰𝗼𝗽𝗲: Agent learns from all users' feedback (trusted teams)
• 𝗨𝘀𝗲𝗿-𝘀𝗰𝗼𝗽𝗲𝗱: Each user gets their own persondocs.weaviate.io/engram?utm_sou…s from their specific usage

Agents can then 𝗹𝗲𝗮𝗿𝗻 𝗳𝗿𝗼𝗺 𝗲𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲 and 𝗶𝗺𝗽𝗿𝗼𝘃𝗲 𝗼𝘃𝗲𝗿 𝘁𝗶𝗺𝗲 both individually and across teams, all kept separate and secure through multi-tenancy.

Find more in the docs: https://t.co/pNyS1JJy0R

💬0🔄0❤️5👀584📊1

⚡ Powered by xgo.ing

A slow search query doesn't 𝘁𝗲𝗹𝗹 𝘆𝗼𝘂 𝘄𝗵𝗮𝘁 𝘁𝗼 𝗼𝗽𝘁𝗶𝗺𝗶𝘇𝗲. The latency could be co...

Weaviate AI Database (@weaviate_io) — Tue, 21 Jul 2026 16:00:52 GMT

A slow search query doesn't 𝘁𝗲𝗹𝗹 𝘆𝗼𝘂 𝘄𝗵𝗮𝘁 𝘁𝗼 𝗼𝗽𝘁𝗶𝗺𝗶𝘇𝗲.

The latency could be coming from filter evaluation, HNSW traversal, compressed-vector rescoring, BM25 scoring, or reading final objects from disk. Each points to a completely different fix.

Weaviate now has 𝗽𝗲𝗿-𝗾𝘂𝗲𝗿𝘆 𝗽𝗿𝗼𝗳𝗶𝗹𝗶𝗻𝗴 to make that breakdown concrete.

Set one opt-in flag on a search and the timing profile comes back directly on 'response.query_profile'. The coordinating node collects profiles from every participating shard and node, so you get a cluster-wide view in one response.

No restart, latency threshold, waiting for the query to be slow again, or manually stitching together logs across nodes.

𝗪𝗵𝗮𝘁 𝘁𝗵𝗲 𝗽𝗿𝗼𝗳𝗶𝗹𝗲 𝗰𝗮𝗻 𝘁𝗲𝗹𝗹 𝘆𝗼𝘂:

- objects_took is high: object hydration is disk-bound. Check page-cache misses, result limits, or large payloads.
- filters_build_allow_list_took is high: the where filter is expensive. filters_ids_matched helps separate a broad cardinality problem from disk reads.
- vector_search_took is high: inspect HNSW layer timings. Layer 0 normally dominates, with ef, dimensionality, and filter strategy as the relevant tuning levers.
- knn_search_rescore_took is high: compressed candidates are expensive to rescore, potentially because full-precision vectors are being read from disk.

For example, one profile might show:

• 48.2ms total
• 36.8ms reading objects
• 8.4ms in vector search
• 2.1ms building the filter allow-list

The vector index isn't the bottleneck. Tuning HNSW would be a guess. The profile points toward storage, page cache, result limit, or payload size instead.

Slow query logging is still useful for passively catching regressions across a fleet. But when one specific query is slow, profiling gives you the actualweaviate.io/blog/query-pro…
Read thdocs.weaviate.io/weaviate/searc…orbach: https://t.co/ygpB4Eb2l8

Docs: https://t.co/70FsH62A3G

💬1🔄4❤️9👀643📊4

⚡ Powered by xgo.ing

𝗠𝗮𝗻𝗮𝗴𝗲𝗱 𝗪𝗲𝗮𝘃𝗶𝗮𝘁𝗲 𝗶𝘀 𝗻𝗼𝘄 𝗮𝘃𝗮𝗶𝗹𝗮𝗯𝗹𝗲 𝗼𝗻 𝗗𝗶𝗴𝗶𝘁𝗮𝗹𝗢𝗰𝗲𝗮𝗻 𝗶𝗻 ...

Weaviate AI Database (@weaviate_io) — Thu, 16 Jul 2026 14:00:29 GMT

𝗠𝗮𝗻𝗮𝗴𝗲𝗱 𝗪𝗲𝗮𝘃𝗶𝗮𝘁𝗲 𝗶𝘀 𝗻𝗼𝘄 𝗮𝘃𝗮𝗶𝗹𝗮𝗯𝗹𝗲 𝗼𝗻 𝗗𝗶𝗴𝗶𝘁𝗮𝗹𝗢𝗰𝗲𝗮𝗻 𝗶𝗻 𝗽𝘂𝗯𝗹𝗶𝗰 𝗽𝗿𝗲𝘃𝗶𝗲𝘄!

DigitalOcean handles the infrastructure around Weaviate:

• Provisioning and security
• Automated daily backups
• Patching and version upgrades
• High availability and storage autoscaling
• Point-in-time cluster forks
• Performance insights and logs in the DigitalOcean Control Panel

It runs the unmodified open-source Weaviate engine (v1.37.1 at launch), so existing Python, JS/TS, Java, and C# clients remain compatible.

Easily spin up a Weaviate cluster without operating idocs.weaviate.io/deploy/install…s or VMs, and get building!

Docs: https://t.co/2uFELXE5k3

💬2🔄3❤️10👀1110📊4

⚡ Powered by xgo.ing

Most AI agents get 𝗱𝘂𝗺𝗯𝗲𝗿 as they learn more. (There's a better way) We just released a demo...

Weaviate AI Database (@weaviate_io) — Tue, 14 Jul 2026 15:00:34 GMT

Most AI agents get 𝗱𝘂𝗺𝗯𝗲𝗿 as they learn more.

(There's a better way)

We just released a demo of 𝗘𝗻𝗴𝗿𝗮𝗺 that shows exactly how intelligent memory management should work.

Tengram.playground.weaviate.io/?utm_source=x&…o/GwVlXXb7Bo

Most agentic chatbots handle memory in one of two broken ways:
• Dump the 𝗲𝗻𝘁𝗶𝗿𝗲 𝗰𝗼𝗻𝘃𝗲𝗿𝘀𝗮𝘁𝗶𝗼𝗻 𝗶𝗻𝘁𝗼 𝗰𝗼𝗻𝘁𝗲𝘅𝘁 every time (problems: cost, degraded performance, no consistency across conversations)
• 𝗦𝘁𝗼𝗿𝗲 𝗲𝘃𝗲𝗿𝘆 𝗺𝗲𝘀𝘀𝗮𝗴𝗲 for retrieval (problems: noise, contradictions, and facts that changed over time)

𝗘𝗻𝗴𝗿𝗮𝗺 takes a completely different approach: 𝗮𝗰𝘁𝗶𝘃𝗲𝗹𝘆 𝗺𝗮𝗶𝗻𝘁𝗮𝗶𝗻𝗶𝗻𝗴 𝗺𝗲𝗺𝗼𝗿𝗶𝗲𝘀 instead of just piling them up.

As conversations happen, Engram runs asynchronous pipelines in the background that:

1. Extract relevant information matching your configured topics
2. Reconcile new memories with existing ones (handling duplicates, preference changes, evolving facts)
3. Store structured, scoped memories in Weaviate for semantic retrieval

This solves the core problem with naive approaches. Rather than cramming ever-growing context into your LLM (causing degradation, latency, and cost issues) or storing raw conversations (which are noisy and contradictory), Engram actively maintains clean, structured memories that can update or be deleted over time.

The result is agentic chatbots that 𝘢𝘤𝘵𝘶𝘢𝘭𝘭𝘺 improve with each interaction, learning from experience without context pollution.

Your browser does not support the video tag.

🔗 View on Twitter

💬1🔄3❤️9👀928📊4

⚡ Powered by xgo.ing

You've typed the perfect search query and gotten completely useless results. Here's what modern pro...

Weaviate AI Database (@weaviate_io) — Thu, 09 Jul 2026 13:50:08 GMT

You've typed the perfect search query and gotten completely useless results.

Here's what modern product discovery looks like when it actually works.

The Weaviate Playground now has an ecommerce demo, and it's a great example of what you can build with vector search.

𝗪𝗲𝗖𝗼𝗺𝗺𝗲𝗿𝗰𝗲 lets you type a keyword, a mood, or even just a vibe - and it routes your query through BM25, semantic vector search, or a hybrid blend automatically. Results come back with real-time faceted filtering by category, brand, and price.

Each product page also surfaces similar items via near-object vector search. That means the same collection is driving both discovery and recommendation - no extra infrastructure needed.

Here's what's going on under the hood:

1. 𝗞𝗲𝘆𝘄𝗼𝗿𝗱 𝗾𝘂𝗲𝗿𝗶𝗲𝘀 → BM25 ranks results by term frequency and relevance
2. 𝗠𝗼𝗼𝗱/𝘃𝗶𝗯𝗲 𝗾𝘂𝗲𝗿𝗶𝗲𝘀 → vector search finds semantically similar products
3. 𝗛𝘆𝗯𝗿𝗶𝗱 → blends both for the best of both worlds
4. 𝗣𝗿𝗼𝗱𝘂𝗰𝘁 𝗽𝗮𝗴𝗲𝘀 → near-object search pulls similar items automatically

Ready to build your own? Copy the prompt, spin it up on the 𝗳𝗿𝗲𝗲 𝘁𝗶𝗲𝗿 in Weaviate Cloud, and go from ecommerce.playground.weaviate.io/?utm_source=ch…rce search in minutes.
See the demo here: https://t.co/4YI9kC4MyF

Your browser does not support the video tag.

🔗 View on Twitter

💬0🔄1❤️11👀1583📊3

⚡ Powered by xgo.ing

Most developers pick the wrong search method. We built a demo so you can see exactly why. Our keyw...

Weaviate AI Database (@weaviate_io) — Wed, 08 Jul 2026 13:50:26 GMT

Most developers pick the wrong search method.

We built a demo so you can see exactly why.

Our keyword vs vector search demo in the Weaviate Playground lets you explore and compare all three search types side-by-side on real datasets.

Here's what each search type does:

𝗕𝗠𝟮𝟱 (𝗞𝗲𝘆𝘄𝗼𝗿𝗱 𝗦𝗲𝗮𝗿𝗰𝗵)
• Matches exact words and phrases using TF-IDF-style scoring
• Rarer terms that appear frequently score higher
• Perfect for names, specific phrases, and precise queries
• But... it misses synonyms and related concepts

𝗩𝗲𝗰𝘁𝗼𝗿 𝗦𝗲𝗮𝗿𝗰𝗵 (𝗦𝗲𝗺𝗮𝗻𝘁𝗶𝗰 𝗦𝗲𝗮𝗿𝗰𝗵)
• Uses dense embeddings to capture meaning
• "Car" and "automobile" match even without shared words
• Understands context and semantic relationships
• But... can sometimes underweight exact lexical matches

𝗛𝘆𝗯𝗿𝗶𝗱 𝗦𝗲𝗮𝗿𝗰𝗵
• Runs both searches in parallel and merges results
• Uses reciprocal rank fusion to combine the best of both
• Gets you strong keyword hits PLUS semantic neighbors
• The best of both worlds for most use cases!

The demo includes multiple datasets - landmarks, movies, science, games, and Pokemon - so you can test queries like "where gladiators used to fight" or "ancient city buried under volcanic ash" and see how each search type performs. Or, simply copy the psearch.playground.weaviate.io/?utm_source=ch…r own data!

See the demo here: https://t.co/UmzVxgGggz

Your browser does not support the video tag.

🔗 View on Twitter

💬3🔄7❤️16👀1317📊10

⚡ Powered by xgo.ing

Most vector database demos show you search. This one shows you how information organizes itself. T...

Weaviate AI Database (@weaviate_io) — Tue, 07 Jul 2026 13:36:03 GMT

Most vector database demos show you search.

This one shows you how information organizes itself.

This newspaper-themed demo in Weaviate Playground showcases vector clustering in action - and it updates every 2 hours with real headlines.

Here's how it works:

𝟭. 𝗛𝗲𝗮𝗱𝗹𝗶𝗻𝗲𝘀 𝗚𝗮𝘁𝗵𝗲𝗿𝗲𝗱
Six news topics polled every 2 hours: general, world, business, technology, sports, and health.

𝟮. 𝗘𝗮𝗰𝗵 𝗦𝘁𝗼𝗿𝘆 𝗨𝗻𝗱𝗲𝗿𝘀𝘁𝗼𝗼𝗱
Every headline becomes a 1,536-dimensional vector capturing its meaning, not just its words.

𝟯. 𝗦𝘁𝗼𝗿𝗲𝗱 𝗶𝗻 𝗪𝗲𝗮𝘃𝗶𝗮𝘁𝗲
Vectors land in a Weaviate collection, searchable by meaning, keyword, or both (hybrid search).

𝟰. 𝗖𝗹𝘂𝘀𝘁𝗲𝗿𝘀 𝗙𝗼𝗿𝗺 & 𝗚𝗲𝘁 𝗡𝗮𝗺𝗲𝗱
The 𝗟𝗲𝗶𝗱𝗲𝗻 𝗮𝗹𝗴𝗼𝗿𝗶𝘁𝗵𝗺 groups similar headlines into communities - these become the actual "stories" underneath. Then an LLM writes a natural label for each cluster.

𝟱. 𝗗𝗲𝗹𝗶𝘃𝗲𝗿𝗲𝗱 𝘁𝗼 𝘁𝗵𝗲 𝗥𝗲𝗮𝗱𝗲𝗿
One card per story. Readers can tune between semantic and keyword search to explore different angles.

This demo combines vector embeddings, hybrid similarity, and graph-based community detection to automatically discover which headlines are really about the same underlying story.

Plus, the demo includes a copy prompt so you can build it yourself 🔥

This is what vector databases were built for: taking unstructured data (news headlines), understanding meaning (embeddings),chronicles.playground.weaviate.io/?utm_source=ch…tering), and making it searchable (hybrid search).

Check it out in Weaviate Playground 💙
https://t.co/gYhQRlvzRq

💬2🔄4❤️12👀1121📊6

⚡ Powered by xgo.ing

Fraud detection used to be slow, rigid, and blind to novel patterns. Here's what conversational dat...

Weaviate AI Database (@weaviate_io) — Thu, 02 Jul 2026 15:00:15 GMT

Fraud detection used to be slow, rigid, and blind to novel patterns.

Here's what conversational data exploration looks like instead.

Powered by the 𝗪𝗲𝗮𝘃𝗶𝗮𝘁𝗲 𝗤𝘂𝗲𝗿𝘆 𝗔𝗴𝗲𝗻𝘁, our fraud detection demo inside the Weaviate Playground shows how natural language questions get turned into structured database queries, complete with filters and aggregations across multiple collections.

Here's what happens under the hood:

1. You ask a plain-English question (e.g., "Which customers have flagged transactions with mismatched shipping addresses?")
2. The Query Agent analyzes your question and determines the exact queries to run
3. It searches across 𝘁𝗵𝗿𝗲𝗲 𝗰𝗼𝗻𝗻𝗲𝗰𝘁𝗲𝗱 𝗰𝗼𝗹𝗹𝗲𝗰𝘁𝗶𝗼𝗻𝘀 simultaneously — Transactions, Customers, and PhoneProducts
4. It applies filters, aggregations, and cross-collection lookups automatically
5. The answer streams back to you in real time

In the demo, you can see the exact queries the Query Agent constructed behind the scenes, and watch how it handles multipart questions that would normally require several manual query steps.

Some of the fraud patterns it surfaces:
• Flagged customers and their transaction history
• Suspicious payment methods
• Mismatched billing and shipping addresses
• Brand-level fraud rates across PhoneProducts

Traditional fraud detection relies on slow, rule-based systems that struggle with novel patterns. This approach lets you explore your data convfrauddetection.playground.weaviate.io/?utm_source=ch…

Try it yourself 👇
https://t.co/pBEiukFD2t

Your browser does not support the video tag.

🔗 View on Twitter

💬1🔄5❤️9👀1183📊5

⚡ Powered by xgo.ing

Sometimes the best way to understand search algorithms is to see them in action on the same data. W...

Weaviate AI Database (@weaviate_io) — Wed, 01 Jul 2026 15:00:06 GMT

Sometimes the best way to understand search algorithms is to see them in action on the same data.

We built an interactive demo in the Weaviate Playground to help you explore the differences between 𝗸𝗲𝘆𝘄𝗼𝗿𝗱, 𝘃𝗲𝗰𝘁𝗼𝗿, and 𝗵𝘆𝗯𝗿𝗶𝗱 𝘀𝗲𝗮𝗿𝗰𝗵 - and actually see how each one returns different results!

Here's what makes each unique:

𝗕𝗠𝟮𝟱 (𝗸𝗲𝘆𝘄𝗼𝗿𝗱) 𝘀𝗲𝗮𝗿𝗰𝗵 matches on exact words and phrases using TF-IDF-style scoring. Rarer terms that appear often score higher. It's excellent for names and precise phrases but misses synonyms.

𝗩𝗲𝗰𝘁𝗼𝗿 (𝘀𝗲𝗺𝗮𝗻𝘁𝗶𝗰) 𝘀𝗲𝗮𝗿𝗰𝗵 uses dense embeddings where similar meanings sit close in high-dimensional space. However, it can underweight exact lexical matches.

𝗛𝘆𝗯𝗿𝗶𝗱 𝘀𝗲𝗮𝗿𝗰𝗵 runs both approaches in parallel and merges the result lists using reciprocal rank fusion or weighted combination. You get strong keyword hits PLUS semantic neighbors in one result set.

The demo gives you different datasets - landmarks, movies, science, games, pokemon - so you can compare all three search types on the same queries and see what each returns.

Or, you search.playground.weaviate.io/?utm_source=ch…t and try it on your own data!

See the demo here: https://t.co/3I6m3y1K38

💬2🔄3❤️25👀852📊8

⚡ Powered by xgo.ing

You searched "cozy fall vibes." The store returned zero results. We built a demo that shows exactly...

Weaviate AI Database (@weaviate_io) — Tue, 30 Jun 2026 15:00:11 GMT

You searched "cozy fall vibes." The store returned zero results.

We built a demo that shows exactly what good search looks like in ecommerce.

It's not just a search bar. Real product discovery needs keywords, semantics, filters, and recommendations all working together.

We built 𝗪𝗲𝗖𝗼𝗺𝗺𝗲𝗿𝗰𝗲 in the Weaviate Playground to show exactly that 🔥

Type a keyword, a mood, or a vibe and it routes your query through:

- 𝗕𝗠𝟮𝟱 keyword search for exact matches
- 𝗩𝗲𝗰𝘁𝗼𝗿 𝘀𝗲𝗮𝗿𝗰𝗵 for semantic, meaning-based results
- 𝗛𝘆𝗯𝗿𝗶𝗱 𝘀𝗲𝗮𝗿𝗰𝗵 for everything in between

Real-time faceted filtering lets you slice results by category, brand, and price instantly. And on every product page, near-object vector search surfaces similar items automatically.

One Weaviate collection powers all of it — search 𝘢𝘯𝘥 recommendations, no extra infrastructure needed.

We have a copy-paste prompt for Cursor, Claude Code, or whatever else you use in our Weaviate Playground so you can easplayground.weaviate.io/?utm_source=ch…Check it out here: https:/ecommerce.playground.weaviate.io/?utm_source=ch…iew the live demo: https://t.co/4YI9kC4MyF

💬2🔄3❤️9👀718📊4

⚡ Powered by xgo.ing

We are so excited to welcome Harneet to our team today! 🎉 Harneet Singh is joining us from Canada ...

Weaviate AI Database (@weaviate_io) — Tue, 30 Jun 2026 11:01:25 GMT

We are so excited to welcome Harneet to our team today! 🎉

Harneet Singh is joining us from Canada as a Security & IT Operations Lead 🧑‍💻

Welcome to Weaviate! 💚

💬2🔄0❤️12👀783📊4

⚡ Powered by xgo.ing

🚀 Weaviate 1.38 is here! Two big capabilities reach general availability in this release — plus a ...

Weaviate AI Database (@weaviate_io) — Thu, 25 Jun 2026 15:00:34 GMT

🚀 Weaviate 1.38 is here!

Two big capabilities reach general availability in this release — plus a faster, simpler path to running replicated clusters at scale.

🧠 HFresh (GA) — our disk-based vector index for billion-scale, constantly-changing data. Low memory, predictable latency, and no periodic full rebuilds.

🤖 MCP Server (GA) — connect LLMs, IDEs, and AI agents directly to Weaviate. In 1.38 you can flip the server (and write access) on or off at runtime — no restart.

🔁 Async replication, everywhere — re-architected to run cluster-wide from a single scheduler, and now on by default for every replicated collection.

⚡ Boost API (preview) — gently promote or demote results at query time, without filtering anything out.

🔎 Nested Object Filtering (preview) — filter on nested object fields with a simple dotted path like cars.make.

…and more: production-ready replica movement, server-side usage guardrails, a new text2vec-digitalocean module, fractional BM25 boosts, and faster startup. 🛠️

👉 Read the full reweaviate.io/blog/weaviate-…vi#Weaviatee#VectorDatabasea#AIa#AIAgentsI#RAGt#MCPA#OpenSourcen#MachineLearningearning

💬3🔄7❤️27👀1785📊9

⚡ Powered by xgo.ing

What happens when your AI remembers you as an engineer, then learns you're CEO? Most systems just s...

Weaviate AI Database (@weaviate_io) — Wed, 24 Jun 2026 14:00:26 GMT

What happens when your AI remembers you as an engineer, then learns you're CEO?

Most systems just store both. Engram does something smarter:

When new information arrives, Engram doesn't just pile it on top of existing memories. It actively reconciles them.

Here's what happens in this case:

1. Engram extracts the new fact: "User has been promoted to CEO"
2. It retrieves related memories from Weaviate, finding "User works as a machine learning engineer"
3. Using an LLM tool call, it determines the appropriate action:
- Rewrite the existing memory: "The user used to work as a machine learning engineer, but has now been promoted to CEO"
- Delete the duplicate new memory to prevent redundancy

The result: no duplicate memories, history is maintained, and the context stays clean.

This is memory as infrastructure. Not just storage, but active maintenance that prevents the ever-growing pile of contradictory facts that would otherwise pollute your agent's context.

Read more about how Engram handles memory reconciliation: weaviate.io/blog/engram-de…

Or get started in the docs: docs.weaviate.io/engram?utm_sou…

💬3🔄2❤️7👀743📊6

⚡ Powered by xgo.ing

Ask "𝘴𝘩𝘰𝘸 𝘮𝘦 𝘤𝘶𝘴𝘵𝘰𝘮𝘦𝘳𝘴 𝘸𝘪𝘵𝘩 𝘮𝘪𝘴𝘮𝘢𝘵𝘤𝘩𝘦𝘥 𝘣𝘪𝘭𝘭𝘪𝘯𝘨 𝘢𝘯𝘥 ...

Weaviate AI Database (@weaviate_io) — Tue, 23 Jun 2026 15:00:44 GMT

Ask "𝘴𝘩𝘰𝘸 𝘮𝘦 𝘤𝘶𝘴𝘵𝘰𝘮𝘦𝘳𝘴 𝘸𝘪𝘵𝘩 𝘮𝘪𝘴𝘮𝘢𝘵𝘤𝘩𝘦𝘥 𝘣𝘪𝘭𝘭𝘪𝘯𝘨 𝘢𝘯𝘥 𝘴𝘩𝘪𝘱𝘱𝘪𝘯𝘨 𝘢𝘥𝘥𝘳𝘦𝘴𝘴𝘦𝘴."

Watch an AI agent write the query, run it, and stream the results back, instantly.

That's exactly what this fraud detection demo in the 𝗪𝗲𝗮𝘃𝗶𝗮𝘁𝗲 𝗣𝗹𝗮𝘆𝗴𝗿𝗼𝘂𝗻𝗱 shows.

Using the 𝗪𝗲𝗮𝘃𝗶𝗮𝘁𝗲 𝗤𝘂𝗲𝗿𝘆 𝗔𝗴𝗲𝗻𝘁, you ask a plain-English question and it automatically translates it into structured queries with filters and aggregations across three different data collections:

- 🗂 𝗧𝗿𝗮𝗻𝘀𝗮𝗰𝘁𝗶𝗼𝗻𝘀
- 👤 𝗖𝘂𝘀𝘁𝗼𝗺𝗲𝗿𝘀
- 📱 𝗣𝗵𝗼𝗻𝗲𝗣𝗿𝗼𝗱𝘂𝗰𝘁𝘀

No SQL. No query builder. Just natural language.

In the demo you'll see:

1. 𝗙𝗹𝗮𝗴𝗴𝗲𝗱 𝗰𝘂𝘀𝘁𝗼𝗺𝗲𝗿𝘀 𝗮𝗻𝗱 𝘀𝘂𝘀𝗽𝗶𝗰𝗶𝗼𝘂𝘀 𝗽𝗮𝘆𝗺𝗲𝗻𝘁 𝗺𝗲𝘁𝗵𝗼𝗱𝘀 identified instantly
2. 𝗠𝗶𝘀𝗺𝗮𝘁𝗰𝗵𝗲𝗱 𝗮𝗱𝗱𝗿𝗲𝘀𝘀𝗲𝘀 and brand-level fraud rates surfaced automatically
3. 𝗠𝘂𝗹𝘁𝗶𝗽𝗮𝗿𝘁 𝗾𝘂𝗲𝘀𝘁𝗶𝗼𝗻𝘀 handled across collections in a single query
4. The exact queries the agent constructed, fully transparent
5. Real-time streamed responses as the agent works

The Query Agent analyses your collection and property descriptions to understand your data, then autonomously determines what to search, what filters to apply and how to aggregate it. It handles multipart questions too. Ask about transactifrauddetection.playground.weaviate.io/?utm_source=ch…uct categories in the same prompt, and the agent breaks it down, runs the appropriate sub-queries, and synthesizes a single coherent response.

Try it yourself in the Weaviate Playground 🔽
https://t.co/AvBmvUMC0J

💬2🔄1❤️7👀731📊4

⚡ Powered by xgo.ing

Most vector database pilots struggle with data ingestion before they even get to search. You build ...

Weaviate AI Database (@weaviate_io) — Thu, 18 Jun 2026 15:00:31 GMT

Most vector database pilots struggle with data ingestion before they even get to search. 
You build a slick retrieval pipeline, it works fine on 1,000 docs… then someone hands you a few million more and you start seeing errors on all sides.

So we decided to cover some of the best practices that could save you a headache or two:

• 𝗦𝗲𝗿𝘃𝗲𝗿-𝘀𝗶𝗱𝗲 �collection.batch.stream://t.co/Fj7K0ObVaM() sets the rate for you; no batch-size tuning, the server handles the import size dynamically. 
• 𝗜𝗱𝗲𝗺𝗽𝗼𝘁𝗲𝗻𝘁 𝗿𝗲𝘁𝗿𝗶𝗲𝘀 — deterministic UUIDs so re-runs never duplicate or double-bill
• 𝗯𝗹𝗼𝗯𝗛𝗮𝘀𝗵 — keep the object embeddings, drop the oversized objects which you are already storing somewhere else (10 TB → a few GB, same search quality)
• 𝗣𝗗𝗙𝘀, 𝗻𝗼 𝗢𝗖𝗥 — just embed the page images with a multimodal model
• 𝗧𝗲𝘅𝘁, 𝗶𝗺𝗮𝗴𝗲, 𝗮𝘂𝗱𝗶𝗼 & 𝘃𝗶𝗱𝗲𝗼 — one collection, one query

Rewritteweaviate.io/blog/data-impo…ree times this quarter? This one's for you 👇

🔗 https://t.co/939ia66tzM

💬0🔄2❤️8👀619📊3

⚡ Powered by xgo.ing

The most requested feature from our community is finally here: Weaviate Cloud access, free ...

Weaviate AI Database (@weaviate_io) — Wed, 17 Jun 2026 16:00:42 GMT

The most requested feature from our community is finally here: Weaviate Cloud access, free 𝘧𝘰𝘳𝘦𝘷𝘦𝘳.

No trial period. No ticking clock. Just full access to Weaviate's vector database, query agent, and memory capabilities, no credit card required.

Whether you're:

- Hacking away on a side project
- Building a proof-of-concept
- Launching a brand-new startup

You can now get started with Weaviate completely free!

A massive thank you to our community for your constant feedback and support. We build for you, and we can't wait to see what you create 💙

Sign uweaviate.io/go/console?utm…BCiRbTGr
Learn more in this blog by @bobvaweaviate.io/blog/weaviate-…67lU19IZ

Your browser does not support the video tag.

🔗 View on Twitter

💬4🔄6❤️32👀2866📊11

⚡ Powered by xgo.ing

Everyone talks about RAG and semantic search. But nobody's showing you how to use 𝗴𝗿𝗮𝗽𝗵 ...

Weaviate AI Database (@weaviate_io) — Tue, 16 Jun 2026 15:00:51 GMT

Everyone talks about RAG and semantic search.

But nobody's showing you how to use 𝗴𝗿𝗮𝗽𝗵 𝗮𝗹𝗴𝗼𝗿𝗶𝘁𝗵𝗺𝘀 + 𝘃𝗲𝗰𝘁𝗼𝗿 𝗲𝗺𝗯𝗲𝗱𝗱𝗶𝗻𝗴𝘀 to cluster real-time data streams:

𝗪𝗲𝗮𝘃𝗶𝗮𝘁𝗲 𝗖𝗵𝗿𝗼𝗻𝗶𝗰𝗹𝗲𝘀 is our live newspaper-themed demo in the Weaviate Playground that showcases the power of vector clustering in action!

Every 2 hours, fresh headlines flood into the Weaview Chronicle. Then, the system automatically clusters them into the underlying stories - articles about a policy change, a tech announcement, or a global event all end up together.

The demo fetches reports, embeds them as vectors, then clusters them using:

- 𝗛𝘆𝗯𝗿𝗶𝗱 𝘀𝗲𝗺𝗮𝗻𝘁𝗶𝗰 𝘀𝗲𝗮𝗿𝗰𝗵 - understanding meaning, not just keywords
- 𝗖𝗵𝗮𝗿𝗮𝗰𝘁𝗲𝗿 𝗻-𝗴𝗿𝗮𝗺 𝘀𝗶𝗺𝗶𝗹𝗮𝗿𝗶𝘁𝘆 - catching textual patterns
- 𝗟𝗲𝗶𝗱𝗲𝗻 𝗰𝗼𝗺𝗺𝘂𝗻𝗶𝘁𝘆 𝗱𝗲𝘁𝗲𝗰𝘁𝗶𝗼𝗻 - a graph algorithm that identifies clusters of densely interconnected nodes (the reports that belong to the same underlying story)

Vector embeddings can do more than just search. The same techniques powering this demo can cluster customer feedback, organize research papers, group support tickets, or structure any stream of unstructured data.

𝗧𝗿𝘆 𝗶𝘁 𝘆𝗼𝘂𝗿𝘀𝗲𝗹𝗳:
The demo includes a 𝗰𝗼𝗽𝘆 𝗽𝗿𝗼𝗺𝗽𝘁 𝗳𝗲𝗮chronicles.playground.weaviate.io/?utm_source=ch…ld your own version! And with Weaviate Cloud's new 𝗳𝗿𝗲𝗲 𝗳𝗼𝗿𝗲𝘃𝗲𝗿 𝘁𝗶𝗲𝗿, there's no barrier to experimenting 💙

View the demo: https://t.co/En6xZZetih

Your browser does not support the video tag.

🔗 View on Twitter

💬7🔄4❤️35👀2047📊16

⚡ Powered by xgo.ing

𝟳 𝘄𝗼𝗿𝗸𝗶𝗻𝗴 𝗱𝗲𝗺𝗼𝘀. From AI memory to fraud detection - all powered by Weaviate. 𝗢𝘂𝗿 ...

Weaviate AI Database (@weaviate_io) — Thu, 11 Jun 2026 12:55:16 GMT

𝟳 𝘄𝗼𝗿𝗸𝗶𝗻𝗴 𝗱𝗲𝗺𝗼𝘀.

From AI memory to fraud detection - all powered by Weaviate.

𝗢𝘂𝗿 𝗻𝗲𝘄 𝗽𝗹𝗮𝘆𝗴𝗿𝗼𝘂𝗻𝗱 𝗶𝘀 𝗻𝗼𝘄 𝗹𝗶𝘃𝗲, with example projects showcasing everything from search, RAG, agents, memory, and more. Every demo comes with copy-paste prompts so you can start building right out of the box!

𝟭. 𝗘𝗻𝗴𝗿𝗮𝗺 - Persistent memory for AI agents. Watch how facts and summaries are extracted in the background and recalled across conversations.

𝟮. 𝗙𝗿𝗮𝘂𝗱 𝗗𝗲𝘁𝗲𝗰𝘁𝗶𝗼𝗻 𝘄𝗶𝘁𝗵 𝗤𝘂𝗲𝗿𝘆 𝗔𝗴𝗲𝗻𝘁 - Ask questions about transactions and products. The Query Agent translates them into structured queries across multiple collections to surface fraud patterns.

𝟯. 𝗪𝗲𝗖𝗼𝗺𝗺𝗲𝗿𝗰𝗲 - Browse, search, and discover products powered by keyword, vector, and hybrid search with real-time faceted filtering and recommendations.

𝟰. 𝗪𝗲𝗮𝘃𝗶𝗲𝘄 𝗖𝗵𝗿𝗼𝗻𝗶𝗰𝗹𝗲𝘀 - A newspaper-themed clustering demo. Reports are embedded and clustered using hybrid semantic search, character n-gram similarity, and Leiden community detection.

𝟱. 𝗩𝗲𝗰𝘁𝗼𝗿 𝗦𝗲𝗮𝗿𝗰𝗵 𝘃𝘀 𝗞𝗲𝘆𝘄𝗼𝗿𝗱 𝗦𝗲𝗮𝗿𝗰𝗵 - Learn the differences between vector and keyword search, and how they compare to hybrid search.

𝟲. 𝗚𝗹𝗼𝘄𝗲 - Skincare recommendation app demonstrating how domain knowledge agents, custom embedding strategies, and vector search create intelligent product recommendations.
playground.weaviate.io/?utm_source=ch…ision-tree agent that picks which tools to run on the fly to query and reason over your Weaviate dweaviate.io/go/console?utm…https://t.co/m3QiEdbI6w
Plus, our new tier means you can starting building FREE forever on Weaviate Cloud. Sign up here: https://t.co/5Hrd8hqkXf

Your browser does not support the video tag.

🔗 View on Twitter

💬1🔄4❤️23👀1158📊8

⚡ Powered by xgo.ing

Most memory systems don’t actually organise memory. They only accumulate it… And that’s where th...

Weaviate AI Database (@weaviate_io) — Wed, 10 Jun 2026 13:36:02 GMT

Most memory systems don’t actually organise memory. 

They only accumulate it… 

And that’s where things start to break.

When memory has no structure, everything gets mixed together:

- One user can affect the wrong memory boundary
- One session can pollute the next
- One use case can contaminate a completely different workflow
- Retrieval gets noisier over time instead of sharper

That’s not memory. That’s blob storage with hope.

Engram takes a stricter approach. It organises memory into **groups**, **topics**, and **scopes**.

- Groups define the use-case boundary by combining the topics you care about with the pipeline that processes them.
- Topics define what kind of information should become memory, because Engram only extracts memories that match your configured topics.
- Scopes define who or what that memory belongs to, including project-wide, user-scoped, and property-scoped boundaries.

So instead of dumping everything into one shared memory pool, you can model memory the way your applications would work.

A coding assistant, for example, could separate:

- Repo-level memory like code architecture
- User-level memory like personal preferences
- Session-level memory like the current conversation summary

Same memory system. Different boundaries. Much cleaner retrieval.

Also, Engram processes raw inputs into maintained memory through async pipelines that extract memories, reconcile them with existing ones, and commit the final results. Weaviate’s multi-tenancy keeps boundaries intact, and the structure keeps retrieval reliable.

Because memory only becomes useful when it has structure.

Learn more: weaviate.io/blog/engram-de…

Get started: console.weaviate.cloud/signin?utm_sou…

💬1🔄5❤️21👀1015📊9

⚡ Powered by xgo.ing

We keep blaming the model for problems caused by bad memory systems. Shoving more chat history into...

Weaviate AI Database (@weaviate_io) — Tue, 09 Jun 2026 13:36:02 GMT

We keep blaming the model for problems caused by bad memory systems.

Shoving more chat history into context is not memory. It’s an invitation to delay, cost, contradiction, and eventually confusion. 

Modern apps need something stricter:

- Extract what facts matter
- Reconcile them with what’s already known
- Remove duplicates and preserve updates
- Retrieve everything reliably later

That’s the idea behind Engram - our managed memory service built on Weaviate.

Instead of treating memory as an ever-growing prompt, Engram treats it as infrastructure. You send raw inputs that could be conversations, strings, or pre-extracted facts, and Engram runs an async pipeline to extract signals, transform them against existing memory, and commit a clean state into the database. 

So when an agent needs to recall something later, it’s not searching a pile of noisy transcripts. It’s retrieving maintained memory.

This matters most when you need agents to:

- Remember user preferences without bloating context
- Learn from prior tasks
- Update stale facts instead of just duplicating them
- Keep memory scoped correctly across users, teams, or workflows

Learn more about Engram: weaviate.io/blog/engram-de…
Get started: console.weaviate.cloud/signin?utm_sou…

💬1🔄3❤️16👀924📊5

⚡ Powered by xgo.ing

Most agent memory systems are just glorified context windows. And this is exactly why production ag...

Weaviate AI Database (@weaviate_io) — Wed, 03 Jun 2026 15:16:58 GMT

Most agent memory systems are just glorified context windows.

And this is exactly why production agents fail at scale.

We've been working on this for months, and it's finally here: 𝗘𝗻𝗴𝗿𝗮𝗺 𝗶𝘀 𝗻𝗼𝘄 𝗚𝗔.

If you've been building agentic applications, you know the problem. Agents that should get smarter over time stay flat instead. They forget user preferences, re-solve the same problems repeatedly, and waste tokens on work that can't be reused. Long context windows help, but cramming them full degrades accuracy, inflates costs, and increases latency.

𝗘𝗻𝗴𝗿𝗮𝗺 𝘀𝗼𝗹𝘃𝗲𝘀 𝘁𝗵𝗶𝘀.

It's a managed memory service built on Weaviate that 𝘢𝘤𝘵𝘪𝘷𝘦𝘭𝘺 𝘮𝘢𝘪𝘯𝘵𝘢𝘪𝘯𝘴 memory instead of just storing it. Asynchronous pipelines extract relevant information from raw data, reconcile it with existing memories (handling deduplication, preference changes, time-evolving facts), and persist clean, structured memory state ready for retrieval.

𝗞𝗲𝘆 𝗰𝗮𝗽𝗮𝗯𝗶𝗹𝗶𝘁𝗶𝗲𝘀:

𝗙𝗶𝗿𝗲-𝗮𝗻𝗱-𝗳𝗼𝗿𝗴𝗲𝘁 𝗔𝗣𝗜 → Add raw data and continue working. Pipelines run asynchronously in the background with durable execution.

𝗧𝗼𝗽𝗶𝗰𝘀 𝗮𝘀 𝗺𝗲𝗺𝗼𝗿𝘆 𝗺𝗮𝗴𝗻𝗲𝘁𝘀 → Natural language descriptions that pull matching information from raw data. You control what's worth remembering.

𝗦𝗰𝗼𝗽𝗲𝘀 𝗳𝗼𝗿 𝗶𝘀𝗼𝗹𝗮𝘁𝗶𝗼𝗻 → Project-wide, user-scoped, or property-scoped memories with hard and soft isolation enforced at the platform level.

𝗖𝗼𝗺𝗽𝗼𝘀𝗮𝗯𝗹𝗲 𝗽𝗶𝗽𝗲𝗹𝗶𝗻𝗲𝘀 → Extract, transform, buffer, and commit steps that manage memories dynamically based on data type and preferences.

𝗕𝘂𝗶𝗹𝘁 𝗼𝗻 𝗪𝗲𝗮𝘃𝗶𝗮𝘁𝗲 → Memory retrieval inherits Weaviate's vector + keyword + metadata search on the same production stack you already trust, using native multi-tenancy to isolate instances.

Whether you're building chatbots that remember user preferences, agents that learn from experience, or multi-agent systems that need shared context, Engram gives you memory as infrastructurweaviate.io/blog/engram-ge…fer, we’re givindocs.weaviate.io/engram?utm_sou…ur first three months of Engram! Sign up before July 15th to claim it. 

Read the blog: https://t.co/24koipQIjO 

Get started: https://t.co/v9fMGMNCej

Your browser does not support the video tag.

🔗 View on Twitter

💬5🔄13❤️41👀3019📊19

⚡ Powered by xgo.ing

The model is the least interesting part of a RAG agent. What actually determines whether an agent s...

Weaviate AI Database (@weaviate_io) — Tue, 02 Jun 2026 15:00:26 GMT

The model is the least interesting part of a RAG agent.

What actually determines whether an agent succeeds in the real world is disciplined system design.

A lot of teams stop once they have a nice convincing demo, but in production, that’s where the hard unwanted problems show up: hallucinations, inconsistent outputs, security gaps, brittle workflows, and agents that no one fully trusts.

Reliable, enterprise-grade RAG agents, on the other hand, are built as complete systems, with a clear structure across four foundational layers:

- Security: Access control, authentication, and data boundaries must be enforced at the platform and retrieval layer. Permissions cannot be “prompted in” without introducing leakage and other risks.

- Retrieval: The agent must consistently retrieve the correct information that is authorized, current, and contextually relevant. This depends on how knowledge is structured, chunked, indexed, and governed, not on the model itself.

- Instructions: Agent instructions should be treated as operating rules. Clear constraints, narrow responsibilities, and predictable output formats are what make agent behavior stable and automatable.

- Guardrails: High-trust agents are designed to fail safely. Responses are grounded in retrieved sources, uncertainty is acknowledged explicitly, and human review is triggered for high-impact decisions.

When these layers work together, RAG agents stop being impressive demos and start becoming systems people actually trust and deploy.

If you’re building or scaling reliable AI agents, our full technical guide with StackAI breaks down the advanced RAG techniques, real workflow architectures, and best practices for security, evaluation, and monitoring.

Download it for free here: stack-ai.com/whitepaper/wea…

💬1🔄2❤️20👀1019📊6

⚡ Powered by xgo.ing

𝘔𝘰𝘳𝘦 𝘨𝘳𝘢𝘯𝘶𝘭𝘢𝘳 𝘳𝘰𝘭𝘦-𝘣𝘢𝘴𝘦𝘥 𝘢𝘤𝘤𝘦𝘴𝘴 𝘤𝘰𝘯𝘵𝘳𝘰𝘭 𝘪𝘯 𝘞𝘦𝘢𝘷𝘪𝘢𝘵𝘦 ...

Weaviate AI Database (@weaviate_io) — Fri, 29 May 2026 14:00:01 GMT

𝘔𝘰𝘳𝘦 𝘨𝘳𝘢𝘯𝘶𝘭𝘢𝘳 𝘳𝘰𝘭𝘦-𝘣𝘢𝘴𝘦𝘥 𝘢𝘤𝘤𝘦𝘴𝘴 𝘤𝘰𝘯𝘵𝘳𝘰𝘭 𝘪𝘯 𝘞𝘦𝘢𝘷𝘪𝘢𝘵𝘦 𝘊𝘭𝘰𝘶𝘥 𝘤𝘰𝘯𝘴𝘰𝘭𝘦.

We've expanded RBAC in the Cloud console with two new roles — 𝗘𝗱𝗶𝘁𝗼𝗿 and 𝗩𝗶𝗲𝘄𝗲𝗿 — joining the existing Owner and Admin roles.

Whether you're a small team or rolling Weaviate out across business units, the principle of least privilege should be the default and not the exception.

Here's how the four roles break down:

→ 𝗢𝘄𝗻𝗲𝗿 - full control, including billing and promoting other Owners
→ 𝗔𝗱𝗺𝗶𝗻 - manages clusters, billing, and day-to-day org activity
→ 𝗘𝗱𝗶𝘁𝗼𝗿 (𝘯𝘦𝘸) - creates, configures, and deletes clusters, without touching billing or user invites; built for engineers shipping on Weaviate day-to-day
→ 𝗩𝗶𝗲𝘄𝗲𝗿 (𝘯𝘦𝘸) - read-only access for security reviewers, finance partners, and other stakeholders who need visibility without edit rights

Enhanced RBAC reduces accidental changes, limits the blast radius of mistakes, and gives platform and security teams the clarity they need to scale confidently. Managing roles takes just a weaviate.io/blog/rbac-over…ion Settings, and changes take effect immediately.

Read the full post 👉 https://t.co/zrHlt5UPBb

💬0🔄3❤️6👀1231📊2

⚡ Powered by xgo.ing

AI agents' vocabulary explained without a 45-minute YouTube video. You're welcome. → MCP → Single ...

Weaviate AI Database (@weaviate_io) — Thu, 28 May 2026 15:00:00 GMT

AI agents' vocabulary explained without a 45-minute YouTube video.
You're welcome.

→ MCP 
→ Single vs. Multi-gent Architecture 
→ Skills 
→ Agentic RAG 
→ Memory

That last one - Memory is a hard one. It's also very easy to get wrong.

So, we've been working on this problem, and it's called Engram - memory and context management built for agents.

Learn more: weaviate.io/blog/engram-de…

💬4🔄6❤️78👀2065📊26

⚡ Powered by xgo.ing

A few weeks ago, @philipvollet, @victorialslocum and @aestheticedwar1 joined the 𝗕𝗶𝗴 ...

Weaviate AI Database (@weaviate_io) — Wed, 27 May 2026 11:31:19 GMT

A few weeks ago, @philipvollet, @victorialslocum and @aestheticedwar1 joined the 𝗕𝗶𝗴 𝗕𝗲𝗿𝗹𝗶𝗻 𝗛𝗮𝗰𝗸 with a mission: build an epic project with Weaviate, and win the top prize.

They only had 36 hours start to finish, and the concept was ambitious - combine trend-dectection on Twitter with persona-based AI content generation.

𝗧𝗵𝗲 𝘁𝗲𝗰𝗵 𝘀𝘁𝗮𝗰𝗸:
• Weaviate for storing and semantically searching Twitter data
• Gemini models for content generation
• Tavily for additional context retrieval
• A custom scoring model built with Pioneer to rank trends

The system monitors Twitter to identify emerging trends using semantic clustering. When it finds something relevant, it analyzes your writing style and previous content to build a persona model. Then it generates new posts about those trends - but in 𝘺𝘰𝘶𝘳 voice, not generic AI-speak.

The question is: did their sleep-deprived hackathon project manage to take home the 10k top prize?

Watchyoutu.be/i1cYfxB8-mw?si…out 👀
https://t.co/Uy1CeQT1TU

💬0🔄4❤️9👀1092📊3

⚡ Powered by xgo.ing

Most companies talk about vector search. Few share what it actually takes to scale to 100M+ embeddi...

Weaviate AI Database (@weaviate_io) — Tue, 26 May 2026 10:56:58 GMT

Most companies talk about vector search.

Few share what it actually takes to scale to 100M+ embeddings in production.

Başak Eskili from @bookingcom joined the Weaviate Podcast to break down their AI journey, and it's packed with insights about what building production systems at massive scale actually looks like.

𝗧𝗵𝗲 𝗘𝘃𝗼𝗹𝘂𝘁𝗶𝗼𝗻:
• Started with keyword matching → semantic retrieval with 𝗢𝗽𝗲𝗻𝗦𝗲𝗮𝗿𝗰𝗵 on AWS
• Scaled to hundreds of millions of embeddings with strict latency requirements
• Migrated to 𝗪𝗲𝗮𝘃𝗶𝗮𝘁𝗲 to handle complex filtering, rising concurrency, and production-scale demands

𝗣𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝗼𝗻 𝗚𝗲𝗻𝗔𝗜 𝗶𝗻 𝗔𝗰𝘁𝗶𝗼𝗻:
Their partner-to-guest messaging agent is a real-world example of 𝗮𝗴𝗲𝗻𝘁𝗶𝗰 𝗔𝗜:
• 𝗪𝗲𝗮𝘃𝗶𝗮𝘁𝗲 retrieves relevant response templates
• 𝗔𝗣𝗜𝘀 fetch property and booking context
• The agent suggests templates, crafts grounded replies, or defers to humans (human-in-the-loop design!)
• Evaluation spans offline datasets, LLM-as-a-judge, A/B testing, and live partner feedback

@CShorten30 and Başak talk about how 𝗕𝗼𝗼𝗸𝗶𝗻𝗴.𝗰𝗼𝗺 tested with 100 million embeddings, filtered vector search, multi-threaded concurrency, reads during writes, and cost-efficient infrastructure provisioning to evaluate Weaviate, as well as a look ahead at personalized travel agents with memory systems that capture user prefereyoutube.com/watch?v=O9edM9…and long-term personalization!

Watch the full podcast here: https://t.co/HB9ceB7JWD

💬0🔄6❤️18👀2314📊7

⚡ Powered by xgo.ing

I asked Claude Code to implement something trivial in my repo. Three turns later, we'd burned 80K to...

Weaviate AI Database (@weaviate_io) — Thu, 21 May 2026 15:00:11 GMT

I asked Claude Code to implement something trivial in my repo. Three turns later, we'd burned 80K tokens because the agent had a hard time finding the right part of the codebase that contained the logic it needed.

Weaviate v1.37.1 ships an MCP server built into the database. Point Claude Code, Cursor, or VS Code at `/v1/mcp` and your agent can ingest your codebase and other materials into Weaviate and perform hybrid search when needed. 

🔍 Hybrid search nails mixed-intent code queries: "Where do we handle retry on 429?" — BM25 anchors on 429 as an exact token, vectors find the semantically related retry code. 

🧩 The MCP server lives inside the database, on the same port as the REST API. No separate process to monitor or external tooling needed.

weaviate.io/blog/coding-as…hQG 

What's the worst context-window failure you've shipped to a coding agent#MCPM#RAGR#VectorSearchr#DeveloperToolsols

💬3🔄7❤️14👀1203📊9

⚡ Powered by xgo.ing

Your vector search just returned five pizzas. You queried "Italian food" and got margherita, marghe...

Weaviate AI Database (@weaviate_io) — Wed, 20 May 2026 15:15:07 GMT

Your vector search just returned five pizzas.

You queried "Italian food" and got margherita, margherita, margherita, margherita, and in a bold twist 𝗺𝗮𝗿𝗴𝗵𝗲𝗿𝗶𝘁𝗮.

All technically correct. All useless together.

This is what happens when relevance is the only thing scoring the results… You get a tight cluster of near-identical objects, ranked by how much they agree with each other.

Weaviate 1.37 ships Maximum Marginal Relevance (MMR). 

One new parameter: 𝘀𝗲𝗹𝗲𝗰𝘁𝗶𝗼𝗻=𝗗𝗶𝘃𝗲𝗿𝘀𝗶𝘁𝘆.𝗠𝗠𝗥(𝗹𝗶𝗺𝗶𝘁=𝟱, 𝗯𝗮𝗹𝗮𝗻𝗰𝗲=𝟬.𝟱), and the algorithm stops letting the fifth result coast on similarity to the first four. It iteratively picks the most relevant item first, then penalises candidates that are too similar to what has already been selected, so each new result has to earn its place by adding something new.

The 𝗯𝗮𝗹𝗮𝗻𝗰𝗲 is the control knob:

- 𝟬.𝟬 = go full throttle and maximise the difference between results.
- 𝟭.𝟬 = don't touch anything, give back standard search results.
- 𝟬.𝟱 = the middle, where each result has to justify itself on both axes.

MMR works across all 𝗻𝗲𝗮𝗿_* queries in Weaviate, and in the query agent it uses 𝗱𝗶𝘃𝗲𝗿𝘀𝗶𝘁𝘆_𝘄𝗲𝗶𝗴𝗵𝘁 for more accurate relevance score from the reranker.

The use cases that get the most benefits are retrieval-heavy agents and standard RAG pipelines. Because if the retrieval step pulls five semantically identical chunks into context, we've essentially wasted four of our five slots. MMR fixes this without changing the schema, index, or collection.

We set a limit at the query level (the candidate pool), then 𝗗𝗶𝘃𝗲𝗿𝘀𝗶𝘁𝘆.𝗠𝗠𝗥(𝗹𝗶𝗺𝗶𝘁=𝟱) for how many objects come back after reranking.

Bigger pool = more material to work with = better diversity.
3-4× yweaviate.io/blog/weaviate-…od starting point.

Learn more about what’s new in the latest release: 

https://t.co/kE2omv8g3r

Five identical pizzas is a solved problem now.

💬2🔄9❤️31👀4091📊12

⚡ Powered by xgo.ing

Stop treating video like text. You don’t need transcripts. You don’t need metadata. You can now em...

Weaviate AI Database (@weaviate_io) — Tue, 19 May 2026 15:00:09 GMT

Stop treating video like text.

You don’t need transcripts.
You don’t need metadata.

You can now embed the videos directly for search.

The pipeline goes very simple:

→ Split raw video into overlapping clips
→ Embed with Gemini embedding 2 multimodal
→ Store in Weaviate
→ Retrieve the exact moments that matter
→ Generate answers grounded in real video

No preprocessing hacks.

You ask a question, it just finds the right moments and answers from them.

Notebook link: github.com/weaviate/recip…

Full multimodal guide: weaviate.io/blog/multimoda…

💬2🔄6❤️21👀1418📊9

⚡ Powered by xgo.ing

A user searches for "caffe crema" in your speciality coffee e-commerce store. The result? 0 matches...

Weaviate AI Database (@weaviate_io) — Fri, 15 May 2026 13:16:48 GMT

A user searches for "caffe crema" in your speciality coffee e-commerce store.

The result? 0 matches.

Your BM25 search just treated "caffé" and "caffe" as two different tokens, and the keyword half of your hybrid search dropped to zero.

Weaviate v1.37 ships three things to improve this:

𝗣𝗲𝗿-𝗽𝗿𝗼𝗽𝗲𝗿𝘁𝘆 𝗮𝗰𝗰𝗲𝗻𝘁 𝗳𝗼𝗹𝗱𝗶𝗻𝗴. One line of schema config and "caffé" matches "caffe" everywhere, at index time and query time.

𝗣𝗲𝗿-𝗽𝗿𝗼𝗽𝗲𝗿𝘁𝘆 𝘀𝘁𝗼𝗽𝘄𝗼𝗿𝗱 𝗽𝗿𝗲𝘀𝗲𝘁𝘀. "The North Face" stops getting destroyed by an aggressive English stopword list. Your French descriptions get a French preset. Same collection.

𝗔 𝗣𝗢𝗦𝗧 /𝘃𝟭/𝘁𝗼𝗸𝗲𝗻𝗶𝘇𝗲 𝗲𝗻𝗱𝗽𝗼𝗶𝗻𝘁. You hand it text + an analyzer config, weaviate.io/blog/tokenizat…tokens BM25 will score.

Learn more in our blog:

https://t.co/OYLEV58hrv

💬0🔄2❤️7👀921📊3

⚡ Powered by xgo.ing

If users don’t know what to ask your chatbot, they’ll leave. Yes, you can add a few static templat...

Weaviate AI Database (@weaviate_io) — Thu, 14 May 2026 14:46:16 GMT

If users don’t know what to ask your chatbot, they’ll leave. 

Yes, you can add a few static templates or orchestrate a whole generation pipeline, but now your simple chat has turned into a complex machine.

But there’s a better way.

Introducing Suggest Queries Mode in the Weaviate Query Agent!

With Suggest Queries Mode, the Query Agent can suggest queries based on the data in your collections. This can help users understand what questions they can ask and make their experience more engaging!

Check out our docs to learn more:  docs.weaviate.io/agents/query/u…

Your browser does not support the video tag.

🔗 View on Twitter

💬2🔄5❤️21👀1399📊7

⚡ Powered by xgo.ing

Spending half your budget on memory just to keep your vector index running? There's a better way. ...

Weaviate AI Database (@weaviate_io) — Wed, 13 May 2026 13:50:57 GMT

Spending half your budget on memory just to keep your vector index running?

There's a better way.

HNSW is the gold standard for vector search, but it needs everything in memory. As datasets grow, that gets expensive.

HFresh flips the model by storing vectors on disk while keeping only a compact centroid index in memory.

𝗛𝗼𝘄 𝗶𝘁 𝘄𝗼𝗿𝗸𝘀
• Divides vectors into small regions called postings
• Uses a compact in-memory HNSW index over centroids to identify relevant regions
• Fetches only the relevant postings from disk for search
• Applies Rotational Quantization at two levels for compression
The result is significantly lower memory usage with predictable latency, even at billion-vector scale

𝗪𝗵𝗲𝗻 𝘁𝗼 𝗰𝗼𝗻𝘀𝗶𝗱𝗲𝗿 𝗛𝗙𝗿𝗲𝘀𝗵
𝗟𝗮𝗿𝗴𝗲 𝗱𝗮𝘁𝗮𝘀𝗲𝘁𝘀 𝘄𝗶𝘁𝗵 𝗵𝗶𝗴𝗵-𝗱𝗶𝗺𝗲𝗻𝘀𝗶𝗼𝗻𝗮𝗹 𝗲𝗺𝗯𝗲𝗱𝗱𝗶𝗻𝗴𝘀 — Memory savings become substantial compared to HNSW
𝗖𝗼𝘀𝘁-𝘀𝗲𝗻𝘀𝗶𝘁𝗶𝘃𝗲 𝗱𝗲𝗽𝗹𝗼𝘆𝗺𝗲𝗻𝘁𝘀 — Run the same workload on smaller infrastructure
𝗪𝗿𝗶𝘁𝗲-𝗵𝗲𝗮𝘃𝘆 𝘄𝗼𝗿𝗸𝗹𝗼𝗮𝗱𝘀 — The cluster-based design avoids the write amplification that HNSW can experience during large imports

HFresh is currently aweaviate.io/blog/weaviate-…we recommend testing in non-productconsole.weaviate.cloud/signin?utm_sou…

Read our blog for more details: https://t.co/lg1jmppEfa

Try HFresh now in Weaviate Cloud https://t.co/wigAn5wXmR

💬0🔄21❤️64👀3342📊22

⚡ Powered by xgo.ing

Your multi-agent RAG system is 𝗰𝗼𝗻𝗳𝗶𝗱𝗲𝗻𝘁𝗹𝘆 𝘄𝗿𝗼𝗻𝗴. And you can't tell by looking at ...

Weaviate AI Database (@weaviate_io) — Tue, 12 May 2026 13:36:02 GMT

Your multi-agent RAG system is 𝗰𝗼𝗻𝗳𝗶𝗱𝗲𝗻𝘁𝗹𝘆 𝘄𝗿𝗼𝗻𝗴.

And you can't tell by looking at the output.

Think about a typical multi-agent RAG pipeline:

1. 𝗥𝗲𝘀𝗲𝗮𝗿𝗰𝗵 𝗮𝗴𝗲𝗻𝘁 retrieves source material from your vector database
2. 𝗦𝘆𝗻𝘁𝗵𝗲𝘀𝗶𝘀 𝗮𝗴𝗲𝗻𝘁 summarizes that material
3. 𝗥𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 𝗮𝗴𝗲𝗻𝘁 draws conclusions from the summary
4. 𝗥𝗲𝘀𝗽𝗼𝗻𝘀𝗲 𝗮𝗴𝗲𝗻𝘁 formats the final output

Now here's the problem:

If the research agent retrieves even ONE low-relevance chunk or stale document, the synthesis agent compresses that flawed content into a confident-sounding summary.

The reasoning agent then treats that summary as established fact.

The response agent presents the conclusion with zero indication that the entire chain rests on a corrupt foundation.

This is what makes 𝗺𝘂𝗹𝘁𝗶-𝗮𝗴𝗲𝗻𝘁 𝗿𝗲𝘁𝗿𝗶𝗲𝘃𝗮𝗹 𝗳𝗮𝗶𝗹𝘂𝗿𝗲𝘀 𝘀𝗼 𝗱𝗮𝗻𝗴𝗲𝗿𝗼𝘂𝘀 - they're invisible at the output layer. The final response looks polished, confident, and completely wrong.

The fix isn't complicated, but it requires intentional design:
• Validate context quality at EVERY retrieval point
• Set relevance thresholds for each agent
• Don't let low-quality context propagate downstream
• Treat each agent's retrieval interface as an independent risk surface

Your LLM is only as good as what it retrieves - and weaviate.io/blog/retrieval… that problem multiplies with every hop.

Learn more in this blog by Devika Ambekar: https://t.co/7TsIaF8XEa

💬4🔄7❤️43👀2027📊18

⚡ Powered by xgo.ing

Everyone's talking about 'agentic AI' but few show you how to build it. This video walks through th...

Weaviate AI Database (@weaviate_io) — Thu, 07 May 2026 15:00:06 GMT

Everyone's talking about 'agentic AI' but few show you how to build it.

This video walks through the complete architecture: Query Agent reasoning, multivector embeddings, specialized collections, and streaming responses with sources. This is how agentic RAG actually works in production.

Legal research demands precision and security and searching across 𝘵𝘰𝘯𝘴 of documents, making it a perfect use case for RAG. But traditionally, building a production-ready legal assistant means months of development time orchestrating retrievers, managing state, and writing complex query logic.

In this video walkthrough, @victorialslocum shows you how to build the entire application from scratch using 𝗪𝗲𝗮𝘃𝗶𝗮𝘁𝗲'𝘀 𝗤𝘂𝗲𝗿𝘆 𝗔𝗴𝗲𝗻𝘁 and the CUAD legal contract dataset.

This isn't just a simple retrieve-then-generate pipeline. Instead, it implements:

𝗠𝘂𝗹𝘁𝗶𝘃𝗲𝗰𝘁𝗼𝗿 𝗣𝗗𝗙 𝗘𝗺𝗯𝗲𝗱𝗱𝗶𝗻𝗴
Instead of OCR and text chunking, the system uses ColModernVBERT with Muvera compression to encode each PDF page directly as visual tokens. This preserves layout, tables, and formatting while keeping retrieval fast.

𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗦𝗲𝗮𝗿𝗰𝗵
The Query Agent doesn't just do keyword matching. It inspects your schema, constructs structured queries with filters, reranks results for precision, and synthesizes grounded answers with source citations. It operates in two modes:

- 𝗦𝗲𝗮𝗿𝗰𝗵 𝗠𝗼𝗱𝗲: Retrieves and reranks relevant contract sections for review
- 𝗔𝘀𝗸 𝗠𝗼𝗱𝗲: Generates direct answers grounded in retrieved context
- streaming responses with cited sources and a full NextJS frontend.

Victoria walks through the complete implementation: dataset setup, PDF ingestion, collection configuration, Query Agent integration, and frontend interyoutu.be/skeKcYbPT9g?si…ed to build your own version.

Watch the video here: https://t.co/ogugAkdmPM

💬2🔄15❤️70👀4831📊25

⚡ Powered by xgo.ing

𝗬𝗼𝘂𝗿 𝗥𝗔𝗚 𝘀𝘆𝘀𝘁𝗲𝗺 𝗽𝗿𝗼𝗱𝘂𝗰𝗲𝘀 "𝗵𝗶𝗴𝗵𝗲𝗿-𝗳𝗹𝘂𝗲𝗻𝗰𝘆 ...

Weaviate AI Database (@weaviate_io) — Wed, 06 May 2026 15:00:07 GMT

𝗬𝗼𝘂𝗿 𝗥𝗔𝗚 𝘀𝘆𝘀𝘁𝗲𝗺 𝗽𝗿𝗼𝗱𝘂𝗰𝗲𝘀 "𝗵𝗶𝗴𝗵𝗲𝗿-𝗳𝗹𝘂𝗲𝗻𝗰𝘆 𝗵𝗮𝗹𝗹𝘂𝗰𝗶𝗻𝗮𝘁𝗶𝗼𝗻𝘀."

More convincing. More confident. More wrong. Here's what research reveals about the real problem.

Devika Ambekar, a PhD candidate at the University of Arkansas researching hallucination detection in multi-agent LLM systems, has found that poor retrieval quality is the single most reliable predictor of degraded output across every pipeline configuration she has studied.

The evidence is clear: when retrieval breaks down, the language model doesn't compensate. It generates with plausible-sounding content that has no grounding in fact.

Her research identifies five critical retrieval failure modes:

1. Retrieval drift (semantically close but contextually insufficient)
2. Context truncation (information silently removed)
3. Stale index poisoning (outdated documents surfacing)
4. Low-relevance top-k retrieval (noise diluting context)
5. Inter-agent miscommunication (failures propagating in multi-agent systems)

Scaling your model doesn't solve a retrieval problem. A more capable LLM given poor context just produces higher-fluency hallucinations.

What builders can do:
• Start with a retrieval audit before upgrading models
• Implement 𝗵𝘆𝗯𝗿𝗶𝗱 𝘀𝗲𝗮𝗿𝗰𝗵 as baseline (dense + BM25)
• Enforce relevance thresholds explicitly
• Track 𝗳𝗮𝗶𝘁𝗵𝗳𝘂𝗹𝗻𝗲𝘀𝘀 as a first-class metric
• In multi-agent systweaviate.io/blog/retrieval…t every retrieval point

Read more in this blog: https://t.co/wgwm05MWCJ

💬0🔄5❤️17👀1478📊8

⚡ Powered by xgo.ing

Remote work. Flexible hours. AI community building. We're hiring 2 people for our growth team at We...

Weaviate AI Database (@weaviate_io) — Tue, 05 May 2026 15:00:15 GMT

Remote work. Flexible hours. AI community building.

We're hiring 2 people for our growth team at Weaviate 👇

𝗗𝗲𝘃𝗲𝗹𝗼𝗽𝗲𝗿 𝗔𝗱𝘃𝗼𝗰𝗮𝘁𝗲 𝗜𝗻𝘁𝗲𝗿𝗻
Help us grow our community by creating engaging technical content, supporting blog posts and videos, and explaining complex AI concepts through appealing visualizations. You'll work with tools like Figma and contribute to our Weaviate Recipes.

𝗗𝗲𝘃𝗲𝗹𝗼𝗽𝗲𝗿 𝗠𝗮𝗿𝗸𝗲𝘁𝗶𝗻𝗴 𝗠𝗮𝗻𝗮𝗴𝗲𝗿
A hybrid role (60% operations, 40% creative) where you'll manage our social channels, create eye-catching content, and bridge the gap between complex AI technology and engaging storytelling for our developer audience.

Both roles are 100% remote with flexible hours, and you'll work at the forefront of AI/ML and cloud-native technologies.

We're looking for someone with enthusiasm for AI, strong writing skills, and a talent for visualization. We're focused on learning together - you don't need to know everything coming in. If you're excited about AI and want to help grow an open-source community, we'd weaviate.io/company/career…
See all open roles here: https://t.co/dPar3y0MUi

Your browser does not support the video tag.

🔗 View on Twitter

💬3🔄4❤️18👀1357📊7

⚡ Powered by xgo.ing

Why would we build a memory product when memory.md already exists? 𝗠𝗘𝗠𝗢𝗥𝗬.𝗺𝗱 ...

Weaviate AI Database (@weaviate_io) — Thu, 30 Apr 2026 16:02:39 GMT

Why would we build a memory product when memory.md already exists?

𝗠𝗘𝗠𝗢𝗥𝗬.𝗺𝗱 𝗿𝗲𝗺𝗲𝗺𝗯𝗲𝗿𝘀 𝘆𝗼𝘂𝗿 𝗰𝗼𝗻𝗰𝗹𝘂𝘀𝗶𝗼𝗻𝘀.

Engram remembers why you made them.

Claude Code's built-in MEMORY.md holds about 200 lines of stable facts: conclusions, preferences, and decisions that don't change session to session. It loads automatically with zero overhead.

But it can't hold everything that led to those conclusions. The reasoning chains, rejected alternatives, the session where the framing shifted doesn't fit in 200 lines and doesn't belong there permanently.

That's where Engram comes in.

Engram structures memory around semantic topics:
• communication-style: output preferences, tone, what to avoid
• domain-context: persistent role and product knowledge
• tool-preferences: languages, frameworks, stack choices
• workflow: how you prefer to work

The workflow runs on strategic triggers:

At session start, Engram recalls with a broad query to prime context before the first question rather than starting cold.

During the session, saves fire on significant moments: a decision made, a direction changed, a deliverable produced. Lightweight saves every few prompts act as insurance against session clearing.

At session end, a full summary gets saved.

Mid-session recall only triggers on specific needs: cross-project references, decision archaeology, resumed work after a gap.

The key insight from building this: letting the LLM decide when to use memory tools doesn't work. Claude defaults to what's already loaded. The integration needs deterministic, infrastructure-level triggers that fire at specific points in the session lifecycle, injecting relevant memories before Claude sees each message.weaviate.io/blog/engram-in…st here: https://t.co/hjzC6Bdgcj

Or sign up fweaviate.io/product-previe…s today! https://t.co/B29pCt6A5A

💬1🔄10❤️27👀1608📊11

⚡ Powered by xgo.ing

What's the difference between AI in demo vs AI in production? 𝗚𝘂𝗮𝗿𝗱𝗿𝗮𝗶𝗹𝘀. Demos show what...

Weaviate AI Database (@weaviate_io) — Wed, 29 Apr 2026 16:02:17 GMT

What's the difference between AI in demo vs AI in production?
𝗚𝘂𝗮𝗿𝗱𝗿𝗮𝗶𝗹𝘀.

Demos show what AI 𝘤𝘢𝘯 do. 

Production systems prove they won't break when things go wrong.

To move from an unpredictable experiment to a reliable enterprise system, you need guardrail implementations. 

Here are the 4 critical patterns in production-grade agentic workflows:

- 𝗔𝗱𝗮𝗽𝘁𝗶𝘃𝗲 𝗙𝗲𝗲𝗱𝗯𝗮𝗰𝗸 𝗟𝗼𝗼𝗽𝘀: The agent evaluates its own output for hallucinations or policy violations. If it fails, it scores the response, generates feedback, and self-corrects in real-time.
- 𝗖𝗼𝗿𝗿𝗲𝗰𝘁𝗶𝘃𝗲 𝗔𝗰𝘁𝗶𝗼𝗻: When an evaluation fails, the system doesn't stop. It triggers automated retries with safer prompts or tighter retrieval filters to ensure the second attempt hits the mark.
- 𝗛𝘂𝗺𝗮𝗻 𝗶𝗻 𝘁𝗵𝗲 𝗟𝗼𝗼𝗽 (𝗛𝗜𝗧𝗟): For high-stakes decisions, the agent pauses and routes the action to a human reviewer. Once approved, the workflow resumes. This balances AI speed with human accountability.
- 𝗘𝗺𝗲𝗿𝗴𝗲𝗻𝗰𝘆 𝗦𝘁𝗼𝗽: If risk thresholds are crossed or the system detects a severe anomaly, it kills the process rather than guessing.

These guardrails give you AI agents that can handle compliance checks, RFP drafting, and claims triage without creating new liabilities.

We just released a complete guide with @stackai covering these pattestack-ai.com/whitepaper/wea…ecture for reliable RAG agents.

Get your free copy here: https://t.co/e2nOYX4cg3

💬0🔄4❤️14👀912📊6

⚡ Powered by xgo.ing

𝗗𝗶𝗴𝗶𝘁𝗮𝗹𝗢𝗰𝗲𝗮𝗻 𝗗𝗲𝗽𝗹𝗼𝘆 '𝟮𝟲 is happening tomorrow in San Francisco, and we're so exc...

Weaviate AI Database (@weaviate_io) — Mon, 27 Apr 2026 16:00:31 GMT

𝗗𝗶𝗴𝗶𝘁𝗮𝗹𝗢𝗰𝗲𝗮𝗻 𝗗𝗲𝗽𝗹𝗼𝘆 '𝟮𝟲 is happening tomorrow in San Francisco, and we're so excited to be part of it!

Deploy is bringing together AI founders, builders, and ecosystem leaders to tackle the biggest challenges of running inference in real-world applications:

• Latency and throughput optimization
• System reliability at scale
• Unit economics that actually work
• Moving from proof-of-concept to production

We're joining industry leaders from 𝗡𝗩𝗜𝗗𝗜𝗔, 𝗩𝗔𝗦𝗧 𝗗𝗮𝘁𝗮, 𝗔𝗿𝗰𝗲𝗲 𝗔𝗜, 𝗖𝗵𝗮𝗿𝗮𝗰𝘁𝗲𝗿 𝗔𝗜, 𝗪𝗼𝗿𝗸𝗮𝘁𝗼, 𝘃𝗟𝗟𝗠, and 𝗜𝗦𝗠𝗚 to share how we're overcoming these production AI hurdles.

These challenges – scaling AI from prototype to production, ensuring reliable performance, and managing costs in AI and agentic applications – are exactly what Weaviate was built to solve, and we’re so excited to partner with @digitalocean to accomplish this 💙

💬2🔄5❤️10👀1810📊5

⚡ Powered by xgo.ing

Weaviate 1.37 is here 🚀 and it makes your database a first-class citizen of the agentic-AI stack wi...

Weaviate AI Database (@weaviate_io) — Thu, 23 Apr 2026 15:00:48 GMT

Weaviate 1.37 is here 🚀 and it makes your database a first-class citizen of the agentic-AI stack with four new preview features:

🤖 MCP Server — A built-in Model Context Protocol server at /v1/mcp. Plug your database directly into Claude Code, Cursor, VS Code, or any MCP-aware tool. 

🌍 Extensible Tokenizers — Keyword search for the non-English world. Accent folding ("Café Crème" → "cafe creme"), custom stopword presets per property, and a /v1/tokenize endpoint that shows exactly what your analyzer is producing.

🎯 Diversity Search (MMR) — Stop retrieving five variations of the same result. A single parameter trades off relevance and diversity at query time. 

⏱️ Query Profiling — Add a timing breakdown to any search request. Zero overhead when disabled. The first tool you should reach for when a query is slow.

Also shipping:

▸ Incremental Backups — unchanged chunks stored as references, dramatically smaller and faster backups for large collections
▸ Gemini Audio — audio joins text, images, and video as a fourth modality in multi2vec-google, powered by the new gemini-embedding-2 model
▸ BlobHash — persist only a SHA-256 hash after vectorization, ideal when the media lives in S3 and shouldn’t be duplicated
▸ Collection Export (preview) — new /v1/export API writes collections to Parquet (S3/GCS/Azure/filesystem)

👉 Full blog pweaviate.io/blog/weaviate-…Xjc#Weaviatev#VectorDatabasea#RAG#MCP#AIAgentsgents

💬4🔄8❤️16👀1349📊10

⚡ Powered by xgo.ing

Your AI agents are burning tokens solving the same problems over and over. Long context windows won...

Weaviate AI Database (@weaviate_io) — Wed, 22 Apr 2026 16:30:14 GMT

Your AI agents are burning tokens solving the same problems over and over.

Long context windows won't save you - here's what will.

Without memory, your agents are stuck in an endless loop of solving the same problems repeatedly, wasting tokens and time. Long context windows seem like a solution, but they degrade accuracy, increase latency, and inflate costs with every request.

Naively storing every message for retrieval creates its own mess. Raw conversations are noisy, contradictory, and full of facts that change over time. You end up with an ever-growing pile of context that degrades rather than improves.

That's why we built 𝗘𝗻𝗴𝗿𝗮𝗺 - a managed memory service running on Weaviate that treats memory as robust infrastructure, not an afterthought.

Here's what makes Engram different:

1️⃣ 𝗜𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝘁 𝗠𝗲𝗺𝗼𝗿𝘆 𝗠𝗮𝗻𝗮𝗴𝗲𝗺𝗲𝗻𝘁

Engram doesn't just store memories - it actively maintains them. When a user tells you they've been promoted, Engram automatically:
• Retrieves related memories about their work
• Rewrites the existing memory to reflect the update while maintaining history
• Deletes duplicate information to keep memories clean

2️⃣ 𝗔𝘀𝘆𝗻𝗰𝗵𝗿𝗼𝗻𝗼𝘂𝘀 𝗣𝗶𝗽𝗲𝗹𝗶𝗻𝗲𝘀

Fire-and-forget your raw data, and let Engram remember what matters. Pipelines extract memories, reconcile new and existing information, and persist to Weaviate - all in the background, so your requests are always low-latency.

3️⃣ 𝗧𝗼𝗽𝗶𝗰𝘀 & 𝗦𝗰𝗼𝗽𝗲𝘀

Topics are "magnets for memories" - natural language descriptions of what information to extract. Scopes control isolation (project-wide, user-scoped, or property-scoped) with hard boundaries enforced by Weaviate's multi-tenancy.

4️⃣ 𝗙𝗹𝗲𝘅𝗶𝗯𝗹𝗲 𝗣𝗶𝗽𝗲𝗹𝗶𝗻𝗲 𝗦𝘁𝗲𝗽𝘀

Extract → Transform → Commit. Engram comes with templates to make getting started simple, but the underlying system is highly configurable. Customize topics, adjust pipeline steps, or take full control of extraction while still leveraging Engram's integration and persistence logic.

5️⃣ 𝗕𝗮𝗰𝗸𝗴𝗿𝗼𝘂𝗻𝗱 𝗧𝗮𝘀𝗸𝘀 𝘄𝗶𝘁𝗵 𝗕𝘂𝗳𝗳𝗲𝗿𝘀

Use buffers to aggregate memories over time or multiple context windows. Engram can extract information from multi-agent systems (input queries, agent actions, user feedback) and combine them into a single useful memory for your agents.

𝗧𝗵𝗲 𝗥𝗲𝘀𝘂𝗹𝘁:

Engram maintains your memory over time so it always stays clean. No duplicate information, historical context preserved, and every memory stays accurate as facts change.

Memory isn't just a featuweaviate.io/blog/engram-de…e. And infrastructure needs to be predictable, scalable, and reliable. 🔥

Read the full deep dive: https://t.co/c9K0SnDawV

💬2🔄14❤️23👀3132📊12

⚡ Powered by xgo.ing

We are so excited to welcome Brandon to our team today! 🎉 Brandon Holwerda - Product Marketing Man...

Weaviate AI Database (@weaviate_io) — Mon, 20 Apr 2026 15:00:30 GMT

We are so excited to welcome Brandon to our team today! 🎉

Brandon Holwerda - Product Marketing Manager 

Welcome to Weaviate 💚

💬0🔄0❤️5👀492📊1

⚡ Powered by xgo.ing

Google Next is 1 week away. Don't leave without stopping by booth #4321 If you're attending Google...

Weaviate AI Database (@weaviate_io) — Thu, 16 Apr 2026 13:23:21 GMT

Google Next is 1 week away.

Don't leave without stopping by booth #4321

If you're attending Google Next from April 22-24th, come say hi to the team! We'd love to connect, chat about what you're building, and share what we've been working on 🥳

💬0🔄1❤️5👀324📊1

⚡ Powered by xgo.ing

Weaviate Shared Cloud is now 𝗴𝗲𝗻𝗲𝗿𝗮𝗹𝗹𝘆 𝗮𝘃𝗮𝗶𝗹𝗮𝗯𝗹𝗲 𝗼𝗻 𝗔𝗪𝗦! Teams standardizing...

Weaviate AI Database (@weaviate_io) — Wed, 15 Apr 2026 15:20:43 GMT

Weaviate Shared Cloud is now 𝗴𝗲𝗻𝗲𝗿𝗮𝗹𝗹𝘆 𝗮𝘃𝗮𝗶𝗹𝗮𝗯𝗹𝗲 𝗼𝗻 𝗔𝗪𝗦!

Teams standardizing on AWS can now run Weaviate in US East (N. Virginia) and Europe (Frankfurt), joining our existing GCP regions. Simply pick the provider and region that best fits your stack.

Shared Cloud is fully managed so that your team can focus on building instead of operations, with automatic cluster upgrades, granular RBAC, immutable backups, and SOC 2/ISO 27001 compliance.

Beyond cluster management, Weaviate Cloud also includes tooling that makes building AI applications meaningfully faster:

→ 𝗪𝗲𝗮𝘃𝗶𝗮𝘁𝗲 𝗘𝗺𝗯𝗲𝗱𝗱𝗶𝗻𝗴𝘀: a native embedding service built into the platform, with no separate API or pipelines to maintain
→ 𝗤𝘂𝗲𝗿𝘆 𝗔𝗴𝗲𝗻𝘁: build RAG out-of-the-box by asking questions about your data in plain language, no database queries required
→ 𝗗𝗮𝘁𝗮 𝗜𝗺𝗽𝗼𝗿𝘁 𝗧𝗼𝗼𝗹: upload files, map fields, and get data flowing into collections in minutes, no code required
→ 𝗗𝗮𝘁𝗮 𝗘𝘅𝗽𝗹𝗼𝗿𝗲𝗿: browse and inspect your data directly in the console

Perfect for teams building on AWS that want to launch AI applications seamlessly on a manaweaviate.io/blog/aws-share…h real platform tooling.

Read the full announcement: https://t.co/97UQ4eLij9

💬0🔄4❤️10👀980📊3

⚡ Powered by xgo.ing

Everyone thinks prompt engineering is just "being nice to ChatGPT". There’s more to it than that. ...

Weaviate AI Database (@weaviate_io) — Tue, 14 Apr 2026 14:22:18 GMT

Everyone thinks prompt engineering is just "being nice to ChatGPT".

There’s more to it than that.

Let's break down the core prompting techniques that actually move the needle with LLMs:

𝗖𝗵𝗮𝗶𝗻 𝗼𝗳 𝗧𝗵𝗼𝘂𝗴𝗵𝘁 (𝗖𝗼𝗧) asks your model to think step-by-step, breaking complex reasoning into intermediate steps. This is especially powerful when working with dense information or conflicting sources in RAG applications. Instead of jumping to conclusions, the model shows its work.

𝗙𝗲𝘄-𝗦𝗵𝗼𝘁 𝗣𝗿𝗼𝗺𝗽𝘁𝗶𝗻𝗴 teaches through examples. By showing the model 2-3 examples of your desired pairs of inputs and outputs, you're essentially pattern-teaching. The model learns how to respond to similar tasks, so that it responds closer in line to what you need it to. Even just one example can be the difference maker in complicated tasks.

𝗧𝗿𝗲𝗲 𝗼𝗳 𝗧𝗵𝗼𝘂𝗴𝗵𝘁𝘀 (𝗧𝗼𝗧) takes CoT further by exploring multiple reasoning paths in parallel - like a decision tree. The model generates several potential solutions and evaluates them to choose the best one. Perfect for scenarios with multiple possible answers or approaches.

𝗥𝗲𝗔𝗰𝘁 𝗣𝗿𝗼𝗺𝗽𝘁𝗶𝗻𝗴 combines reasoning with action in an agentic loop. The model alternates between thinking (Reason) and doing (Act), creating a Thought-Action-Observation cycle. It can interact with tools, observe outcomes, and adjust its reasoning iteratively.

Here's the insight: 𝗲𝗮𝗰𝗵 𝘁𝗲𝗰𝗵𝗻𝗶𝗾𝘂𝗲 𝗮𝗱𝗱𝗿𝗲𝘀𝘀𝗲𝘀 𝗮 𝗱𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝘁 𝗰𝗵𝗮𝗹𝗹𝗲𝗻𝗴𝗲. CoT structures reasoning, Few-shot guides format and style, ToT explores alternatives, and ReAct enables dynamic interaction.

So why pick just one?

Combining 𝗖𝗼𝗧 + 𝗙𝗲𝘄-𝘀𝗵𝗼𝘁 gives you structured reasoning with clear output format. Adding 𝗧𝗼𝗧 lets you explore multiple solutions. Layering in 𝗥𝗲𝗔𝗰𝘁 makes your agent adaptive and tool-aware.

The most effective prompts don't rely on a single technique - they strategically combistack-ai.com/whitepaper/wea…what the task actually needs.

Read about prompting techniques and more in our ebook with @stackai: https://t.co/e2nOYX4cg3

💬0🔄8❤️19👀1183📊8

⚡ Powered by xgo.ing

Production-ready legal AI within 24 hours. Here's the technical breakdown of how we did it for our ...

Weaviate AI Database (@weaviate_io) — Thu, 09 Apr 2026 15:24:25 GMT

Production-ready legal AI within 24 hours.

Here's the technical breakdown of how we did it for our own finance team.

When our finance team asked for help navigating internal contracts, we used Weaviate's 𝗤𝘂𝗲𝗿𝘆 𝗔𝗴𝗲𝗻𝘁 to build a production-ready legal assistant in just a day and a half. Not a demo - a real application with advanced embeddings, agentic search, streaming responses, and cited sources.

Legal research demands extreme precision. Without a reasoning layer, traditional systems lack the ability to apply proper filters before searching.

𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝘀𝗲𝗮𝗿𝗰𝗵 changes this entirely. The Query Agent treats your database as a set of tools rather than a static data store. It inspects your schema, constructs structured queries with the right filters, reranks results for actual relevance, and synthesizes grounded answers with citations.

The architecture we used:

𝗗𝗮𝘁𝗮 𝗹𝗮𝘆𝗲𝗿: PDFs embedded with a multivector model that encodes each page as visual tokens, preserving layout and tables. Muvera compression then reduces memory and latency while maintaining retrieval quality.

𝗦𝗰𝗵𝗲𝗺𝗮 𝗱𝗲𝘀𝗶𝗴𝗻: Three collections instead of one - Commercial Agreements, Corporate & IP Agreements, and Operational Agreements. This gives the agent explicit structure to reason about and route questions effectively.

𝗤𝘂𝗲𝗿𝘆 𝗔𝗴𝗲𝗻𝘁: Operates in Search Mode for discovery and retrieval, or Ask Mode for synthesized answers. Every response includes cited source passages to reduce hallucinations.

With 𝗪𝗲𝗮𝘃𝗶𝗮𝘁𝗲 𝗔𝗴𝗲𝗻𝘁 𝗦𝗸𝗶𝗹𝗹𝘀, you can now build this entire application with a single prompt in Claude Code or Cursor.

Install Agent Skills:
npx skills add weaviate/agent-skills

Then run the prompt (detailed in our blog post) that instructs the coding agent to download the CUAD legal contract dataset, set up the three collections, configure multivector embeddings, and build the frontend interface with source citations.

The agent handles everything: data processing, schema creation, embedding configuration, Query Agent setup, and frontend development. From my tests, it takes roughly 12 minutes and about 20k tokens on Sonnet 4.6.

This same approach works for any document-heavy domain requiring precision: compliance documentation, medical records, technical speweaviate.io/blog/legal-rag…ial analysis.

Read the full technical breweaviate.io/blog/weaviate-…OJCPX8Ps
Dive deeper into Weaviate Agent Skills:  https://t.co/x6lg7EWhw5

💬2🔄8❤️37👀1364📊14

⚡ Powered by xgo.ing

We spent weeks testing text vs. image retrieval for RAG. The winner? 𝗡𝗲𝗶𝘁𝗵𝗲𝗿. Our recent pu...

Weaviate AI Database (@weaviate_io) — Wed, 08 Apr 2026 15:13:51 GMT

We spent weeks testing text vs. image retrieval for RAG.

The winner? 𝗡𝗲𝗶𝘁𝗵𝗲𝗿.

Our recent publication, IRPAPERS, compares 𝘁𝗲𝘅𝘁-𝗯𝗮𝘀𝗲𝗱 𝗿𝗲𝘁𝗿𝗶𝗲𝘃𝗮𝗹 (OCR + vector, keyword, and hybrid search) and 𝗶𝗺𝗮𝗴𝗲-𝗯𝗮𝘀𝗲𝗱 𝗿𝗲𝘁𝗿𝗶𝗲𝘃𝗮𝗹 (multimodal late interaction with ColModernVBERT) to see which one ranks best for PDF search tasks.

The benchmark tested retrieval over 3,230 pages from 166 scientific papers using 180 needle-in-the-haystack queries. Here's what we found:

𝗧𝗲𝘅𝘁-𝗯𝗮𝘀𝗲𝗱 𝗿𝗲𝘁𝗿𝗶𝗲𝘃𝗮𝗹 (Arctic 2.0 + BM25 hybrid search):
• 46% Recall@1
• 78% Recall@5
• 91% Recall@20

𝗜𝗺𝗮𝗴𝗲-𝗯𝗮𝘀𝗲𝗱 𝗿𝗲𝘁𝗿𝗶𝗲𝘃𝗮𝗹 (ColModernVBERT multi-vector embeddings):
• 43% Recall@1
• 78% Recall@5
• 93% Recall@20

Text edges out images at the top rank, but images match or exceed text at deeper recall levels.

But the most interesting thing we found is that 𝘁𝗵𝗲𝘀𝗲 𝗮𝗽𝗽𝗿𝗼𝗮𝗰𝗵𝗲𝘀 𝗳𝗮𝗶𝗹 𝗼𝗻 𝗱𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝘁 𝗾𝘂𝗲𝗿𝗶𝗲𝘀.

At Recall@1:
• 22 queries succeeded with text but failed with images
• 18 queries succeeded with images but failed with text

This means neither 𝘢𝘭𝘰𝘯𝘦 is necessarily better than the other. Text excels at lexical precision and top-rank accuracy. Images preserve visual structure and spatial relationships that text transcription loses.

𝗠𝘂𝗹𝘁𝗶𝗺𝗼𝗱𝗮𝗹 𝗵𝘆𝗯𝗿𝗶𝗱 𝘀𝗲𝗮𝗿𝗰𝗵 fuses both text- and image-based methods together, giving the best results overall:
• 49% Recall@1 (+3 points over text alone)
• 81% Recall@5
• 95% Recall@20

Hyyoutube.com/watch?v=lgzkLw…he best of both worlds - where one method fails, the other might succeed.

Watch the full video by @victorialslocum here: https://t.co/beEYUotMKi

Your browser does not support the video tag.

🔗 View on Twitter

💬2🔄13❤️73👀5536📊23

⚡ Powered by xgo.ing

PDF import just landed in Weaviate Agent Skills! Point Claude Code (or any agent) at a PDF, and it ...

Weaviate AI Database (@weaviate_io) — Tue, 07 Apr 2026 16:30:21 GMT

PDF import just landed in Weaviate Agent Skills!

Point Claude Code (or any agent) at a PDF, and it handles everything from schema setup, collection configuration, to embedding each page using ColModernVBERT multimodal model.

MUVERA then makes multi-vector retrieval efficient by reducing memory usage and computational cost while preserving retrieval quality.

In short: your entire document library becomes searchable in Weaviate with a single prompt. 

CSV, JSON, and JSONL are already supported!

Try it: github.com/weaviate/agent…

Your browser does not support the video tag.

🔗 View on Twitter

💬1🔄5❤️15👀868📊5

⚡ Powered by xgo.ing

We are so excited to welcome Pierre to our team today! 🎉 Pierre Demagny is joining us from France ...

Weaviate AI Database (@weaviate_io) — Tue, 07 Apr 2026 14:30:11 GMT

We are so excited to welcome Pierre to our team today! 🎉

Pierre Demagny is joining us from France as a Cloud Platform Engineer 🧑‍💻

Welcome to Weaviate! 💚

💬1🔄0❤️3👀391📊2