Elastic

Still writing Lucene queries in your Grafana dashboards? The Elasticsearch plugin now supports ES|QL. One feature flag in Grafana 13.0 (elasticsearchESQLQuery = true) and you get a Monaco editor with syntax highlighting, autocompletion, and inline errors. Pipe-based queries against your logs, metrics, and traces without leaving the Grafana UI. The TS command is worth trying first: time series aggregation functions like RATE(), AVG_OVER_TIME(), and DELTA() run through ES|QL's vectorized compute engine. Internal benchmarks suggest significant order-of-magnitude gains over equivalent Query DSL in TSDS data Built by Elastic with Grafana Labs and contributed upstream to the open source project. Full walkthrough with example queries for logs, metrics, and time series: go.es.io/3Psg3t7

Full breakdown: go.es.io/49D2IoL

Stop asking AI to write its own skills Skills encode expertise the model doesn't have. So asking an LLM to generate its own skill is like getting a new grad to write their own onboarding. Better approach: start from real failures, write evaluation tasks first, then build the skill to improve the score.

Learn more: go.es.io/4dH9cFs

If you've ever hit the context window limit after feeding large documents into an LLM, here's what you can do: Override the chunking settings in your semantic_text field, picking recursive chunking. Fixed-size splitting via words and sentences is fast and predicable, but it can split related elements. Recursive chunking hits a better balance, especially for unstructured and semi-structured text: • Tries large separators first (paragraphs → sentences → words) • Falls back gracefully when chunks exceed your size limit (via the • Keeps related content together It won't magically fix a messy corpus, but it's a pragmatic starting point before reaching for semantic or agentic chunking.

Full benchmark methodology, reproduction code, and the math behind centroid centering vs. Hadamard rotation: go.es.io/4dH960y

TurboQuant does win on reconstruction MSE. But most of that advantage is the Hadamard rotation, not the Lloyd-Max centroids. We applied the same rotation to OSQ and the gap nearly vanishes — 0.306 vs 0.307 at 1-bit. More importantly: MSE measures reconstruction quality. Search engines rank by dot products. Those are different objectives, and the gap between them is where our data-dependent design pays off.

🧵 Our BBQ at 1-bit/doc beats TurboQuant at 4-bit/doc on shifted data on ranking accuracy. At 1/5 the storage. We center on the segment centroid before quantization, so the bits go where they are actually needed for ranking. TurboQuant's Hadamard rotation can't exploit that structure.

Clear breakdown of jina-embeddings-v5-omni The diagram alone is worth the click. 👇

Ever had an unmapped field buried in _source that you couldn't query without reindexing the whole index? That's the mapping cliff edge. Field wasn't anticipated at ingest time, so it's invisible to queries. Traditional fix: update the mapping, reindex terabytes of data, wait hours. ES|QL now has a one-line fix: SET unmapped_fields="load" That's it. Every unmapped field in _source becomes queryable at query time. No reindex. No pipeline change. Works retroactively against data you ingested months ago. The trade-off: unmapped fields skip the inverted index, so filters on them are slower than the mapped fast path. But "slower and accessible right now" beats "fast and doesn't exist" when you're debugging at 2am. Pair it with JSON_EXTRACT for surgical extraction from raw JSON strings or flattened fields where dot notation can't reach. Mapped fields stay fast. Unmapped fields stop being invisible. You pick per field, per query. Read more: go.es.io/3PnJfkZ

Building a time series in Kibana Dashboards used to take 14 clicks. Now it takes three. • Brand new API • Natural language dashboard creation • Plus much much more

Combine them with Jina AI’s v5 text embedding models, and you can have a competent RAG pipeline for around 1B parameters, or an even better one for only 5-10B parameters. What’s even better is that these Gemma and Qwen models are multimodal, so they can understand images as well as text. You can combine these with Elasticsearch (running on your machine or in the cloud) and you’ll have an efficient RAG pipeline at a fraction of the compute you used to need. Try them out: go.es.io/4dlDbld

🧵 A competent RAG pipeline with a billion parameters? It’s no longer just a pipe dream. The recent release of Gemma and Qwen models give developers access to models as small as under 1B parameters that can run on not just your computer, but edge devices.

Read the blog: go.es.io/42ZRZRt

jina-embeddings-v5-omni is now on Elastic Inference Service. Text, images, audio, video. One index, one query. • Best-in-class visual understanding under 1B parameters • Beats models 20x its size on multilingual visual tasks • Beats ByteDance Seed 1.6 on video (55.57 vs 29.30 on Charades-STA) • BBQ quantization: 93% storage reduction, under 3% accuracy loss • nano runs on commodity hardware without GPU Introduction below with @florianhoenicke

Read more about Elastic 9.4 and Workflows (now GA): go.es.io/4ddTBvW

Yeah your agent can reason over data. But can it do anything about it? Workflows is the action layer your agents are missing: • Connect to external systems natively • Run multistep sequences automatically • Coordinate across tools without leaving Elastic Agent Builder = what the agent knows. Workflows = what the agent does. New: more connectors, more steps, and natural language authoring in tech preview. One less reason to bolt on a separate orchestration tool.

Great work from the @elasticseclabs team!