Flux: Find What’s Next, Now

Flux is a web search engine.

Existing engines are good at organising the web, but many of their indexing and retrieval methods were never designed for the volume and speed of information creation today, particularly when it comes to time. We think of legacy search, which returns a ranked list of links, as 2D search.

Information Has a Lifespan

Information changes quickly. It loses and gains relevance in hours, and the truth constantly mutates as billions of people measure, comment on, and share their version of reality. Some information is evergreen, like a birthdate. Other information updates rapidly, like breaking news. Treating both the same way is where most search falls down.

The Agentic Shift (3D)

Agentic searches will soon outnumber human ones. LLMs can use tools to issue complex, numerous queries and process the results far faster than any person. We call this 3D search.

The trade-off is that AI easily loses track of where information sits in time. That is how you end up with mistakes like naming the wrong current Prime Minister, or citing medical advice long after it stopped being accurate. The problem only compounds as token costs push us to feed models smaller and smaller fragments of context, eroding the precision that temporal awareness depends on.

A New Way to Index (4D)

So Flux proposes a different way to index. We are over a year into actively building our own search indices, and we do two things.

First, we continuously evolve our crawl infrastructure to gather information as fast as possible, so the index stays on top of what is current.

Second, we organise web data by time. We capture whether a fact is evergreen, like a birthdate, or true only for a given period. We track its duration, its expected expiration, and when an update or change was likely, which enriches cross-checking and verification.

Granular by Design

We do this at the fragment level, so both humans and AI have context for even the smallest pieces of a web page, often 20-100x smaller than a full document.

That provenance also helps with agentic harnesses and tracing. As these flows grow more complex, knowing exactly when and where each fragment came from becomes essential to debugging and trust.

In this way, Flux aims to build 4D search: a way to organise the past and find what’s next, now.