Grounded web search for AI agents
Gather concise and fresh web data with sentence-level citations.
sodium-ion batteries 2026 mass production
Sodium-ion battery momentum grows, but challenges remain
CATL, the world’s largest battery manufacturer, released its second-generation sodium-ion batteries and has confirmed plans for their commercial-scale deployment across multiple sectors starting in 2026.
How CATL hit GWh sodium-ion battery scale
The Naxtra sodium-ion battery achieved GWh-scale industrialization with full production starting in 2026
How Sodium-Ion Technology Is Disrupting the Global Battery Market ...
CATL is extending these batteries to additional passenger vehicles in Q2 2026 and to commercial vehicles in Q3 2026.
Beyond lithium: how sodium-ion batteries could change the world
Batteries based on sodium ions are entering mass production.
CATL to mass-produce Sodium-Ion Batteries in 2026, targets 600 km range
CATL Chief Scientist Wu Kai confirms large-scale mass production of sodium-ion batteries for 2026 and outlines next-generation lithium-air ...
- Page metadatapublished_at 2026-11-19Resolved dates in the textFull results come out Nov. 30Nov 30, 2026The first patients were dosed two weeks agoNov 5, 2026
Fresh
Searches run on the live web for maximum freshness, and every result comes with its citation and publish date.
- [1] digital-strategy.ec.europa.eu
Cited
Every passage links back to the page it came from.
- $4 / 1k
Page reads included
$4 per 1,000 searches, with the pages read for you. Failed requests are free.
- Opt-inZero data retention
Private
Encrypted in transit and at rest, by default. Turn on zero data retention in settings, or ask us for data residency.
Your model's new favourite tools
Lost context is written back in
Lineage names the who, what and when in every passage, so agents don't confuse who did what, even across hundreds of results.
A chunk on its own
With Lineage
Resolves pronouns and unnamed entities
Theirwomen who received sac-TMT overall risk of death was also 32% lower.
news.cancerresearchuk.org
Resolves vague references
CATL describes the agreementCATL and HyperStrong’s 60-GWh sodium-ion agreement as the largest sodium-ion battery supply deal in the world to date, and HyperStrong as the company's first strategic partner for sodium-ion energy storage.
- organisationCATL
- organisationHyperStrong
batterytechonline.com
Completes partial dates
Speaking at the 2026 Equipment Powerhouse Forum on May 30May 30, 2026 (year inferred), Wu Kai, Chief Scientist at CATL and Academician of the Chinese Academy of Engineering, confirmed that the manufacturing bottlenecks have been resolved, according to Sina.
- personWu Kai
- organisationCATL
- organisationSina
carnewschina.com
Accurate answers from a fraction of the context
Lineage keeps most of the answer quality while cutting the context bloat that confuses models on long-horizon tasks.
8.4x
fewer tokens per question
597, down from 5,018 for the whole page.
90%
of the whole page's answer quality, with 88% fewer input tokens
0.418 F1 against 0.465.
27x
fewer tokens with 16 sources
260, down from 7,025 for whole pages.
400 questions on research papers (QASPER).
Answer F1, higher is betterContext tokens per question, log scale
| Method | Context tokens per question, log scale | F1 |
|---|---|---|
| Lineage | 597 | 0.418 |
| Whole page | 5018 | 0.465 |
| Keyword retrieval | 640 | 0.308 |
| Truncation | 623 | 0.117 |
When the answer sits on only one of 16 sources, Lineage returns next to nothing from the rest, so your model doesn't read irrelevant content.
Whole pageLineage
| 1 | 605 | 245 |
| 4 | 1899 | 250 |
| 8 | 3609 | 256 |
| 16 | 7025 | 260 |
Deep research that reads deeply. It asks follow-up questions in parallel and reads around 500 sources in about 20 seconds, finding answers in pages you wouldn't usually consider. Only the passages that answer reach your model, so every extra person, company or technology it finds costs little, and your cost per task stays low.
0sources read
What your model reads
Whole page
Lineage
Tested with gpt-oss-120b on 400 questions per method and scored against reference answers. Whole pages score 0.465 F1 and Lineage 0.418, with 8.4x fewer tokens. Short pages lose more: on Quoref with one source, F1 drops from 0.851 to 0.651.
Build on Flux
Get started
- 01
Get a key
Sign up and create a key. You get free credit to start.
Create a key - 02
Make your first request
Call Search, Lineage or Fetch with the same key.
Read the docs - 03
Connect your agent
Give your agent fluxsearch.io/llms.txt and it can learn the whole API.
Open llms.txt
Only pay for what you use
Usage-based billing.
Questions
What is Flux?
An API that lets AI agents search the web and read pages. Search finds pages for a question, Lineage pulls out the sentences that answer it, and Fetch gives you a whole page as markdown. One key works for all three.
When should I use Search, Lineage or Fetch?
Search when you don't have sources yet. Lineage when you have a page and only want the answer. Fetch when you want the whole page.
How do citations work?
Every passage comes with its link and date, and the people, companies, places and dates in it are tagged. In Flux chat, each claim links to the passage it came from.
How fresh are the results?
Search runs on the live web when you call it, and shows each page's publish date where there is one.
What does it cost to start?
Nothing up front. New accounts get free credit. After that, pay as you go or pick a monthly plan with bonus credit.
What happens to the data I send?
Requests are encrypted in transit and at rest. You can turn on zero data retention in settings, and we offer data residency on request.