About FlatSearch

We didn't build an API company.

We built a financial research platform — and FlatSearch came out of the wreckage of trying to power it with everyone else's search tools.

Our core product is VCP Scanner, a financial equity research platform that tracks, analyses, and models over 5,800 US-listed stocks every single day. To power the agents behind it, we needed to ingest massive amounts of live web context continuously — earnings data, analyst sentiment, regulatory filings, news — all synthesised and structured, at speed, at scale.

We tried the obvious options first.

Serper gave us raw links. We'd have to build our own scraping layer on top, manage CAPTCHAs, clean the output ourselves. That was a full engineering project just to get to the starting line.

Tavily and Perplexity looked better until we actually ran them at volume. Variable token pricing hit us hard. A single deep search query would silently cost 5x the advertised base rate depending on what pages it hit. When you're running thousands of queries a day across 5,800 stocks, that variance destroys your margin.

So we stopped buying and started building.

We wired residential proxy clusters to bypass CAPTCHA walls, built aggressive boilerplate-stripping pipelines to clean raw HTML before it touched an LLM, and connected everything to frontier synthesis models for fast, cheap, structured output. We built what we needed because nothing we could buy did the job at our cost constraints.

That infrastructure now runs 24/7. It has to — market hours are relentless.

Why we opened it up

Our proxy clusters and server infrastructure run at roughly 40% utilisation during off-market hours. We're paying for that capacity regardless.

Wrapping our internal pipeline into a REST API and offering it to other developers at our wholesale cost basis was the obvious move. We already despise variable token pricing — we got burned by it ourselves. So every FlatSearch endpoint is a strict flat rate. If a query takes our agents three attempts to get right, we absorb the cost. Not you.

You are not buying a product someone built to sell you. You are renting capacity on infrastructure another business depends on to operate.

Battle-tested means something specific here

This isn't a weekend wrapper project. The exact pipeline behind FlatSearch runs financial analysis across thousands of equities every trading day. Uptime isn't optional for us. Neither is output quality — bad synthesis on a stock research query is a real problem with real consequences.

When we say production-grade, we mean it in the context of a business that cannot afford for it not to be.