EN
Back to the archive

The archive · Developer & Business Tools · Product decision · 2025–2026

Nozomio bets context, not generation, is coding agents' bottleneck; $6.2M seed

Solo teen founder Arlan Rakhmetzhanov sells Nia, a context layer for coding agents — 90+ paying YC startups, $11.5K MRR, then a $6.2M CRV-led seed.

Nozomio

The betThat context, not code generation, is the bottleneck — and before model providers settle it, developers will pay for a layer that indexes docs and codebases into Cursor.Live

What the business is

Nozomio sells Nia, an MCP server and API that indexes entire codebases and documentation sites and feeds clean context to coding agents (Cursor, Continue, Cline), plus a deep-research agent that compares libraries.

Starting capital$1M pre-seed raised in days from LocalGlobe in London, then a $6.2M seed (announced Oct 2025) led by CRV with BoxGroup, LocalGlobe and Y Combinator, plus Paul Graham, Gustav Söderström (Spotify CTO) and Thomas Wolf (Hugging Face).

How it started

Arlan Rakhmetzhanov, a 17-year-old from Kazakhstan who dropped out of high school and moved to London, raised a $1M pre-seed from LocalGlobe in days and got into YC on his third attempt. After his own coding-agent experiments failed, he concluded agents were bad at retrieving the right context, and launched Nia through YC's Summer 2025 batch.

What happened

Within three weeks of outreach, 25% of the current YC batch and 5% of the prior spring batch were paying — 90+ companies and $11.5K MRR with unit economics already profitable (Sept 2025). Just before Demo Day he closed a larger round; tech.az reported the $6.2M seed on Oct 3, 2025. By Aug 2026, Nia's Launch YC post (221 upvotes) claimed a 27% improvement in Cursor's performance on internal evals.

No ending yet — it is still running.

Background

Arlan Rakhmetzhanov, a teenage dropout from Kazakhstan, moved to London, raised a $1M pre-seed from LocalGlobe in days, and got into YC on his third attempt. After his own coding-agent experiments failed, he concluded that generation was no longer the bottleneck — retrieval was: agents were great at writing code and terrible at getting the right context.

Nozomio sells Nia, an MCP server and API that indexes entire codebases and documentation sites so agents like Cursor, Continue and Cline 'just know' the right answer. Within three weeks of outreach, 25% of the current YC batch and 5% of the prior spring batch were paying — 90+ companies, $11.5K MRR, and unit economics already profitable (Sept 2025).

Just before Demo Day the founder closed a larger round; tech.az reported a $6.2M seed led by CRV with BoxGroup and LocalGlobe on Oct 3, 2025, backed by Paul Graham, Spotify's CTO and Hugging Face's co-founder. By Aug 2026, Nia's Launch YC post (221 upvotes) claimed a 27% improvement in Cursor's performance on internal evals — an early bet that context infrastructure is a layer worth paying for.

What has to be true

  • The founder attacked a felt problem — context pollution and lost sessions — rather than building another coding agent, so the pain was already real for every Cursor user (TBPN).
  • Dogfooding YC as a market: batchmates were the densest coding-agent users, giving fast feedback, fast revenue and a credible reference list (TBPN).
  • Timing helped: MCP had just standardized the integration surface, letting a three-person team plug into Cursor, Continue and Cline without building IDEs (Launch YC).
  • Profitable unit economics meant the company did not depend on the next round while the RAG-versus-longer-context debate plays out (TBPN).
  • A young solo founder with Paul Graham, Spotify and Hugging Face names on the cap table made a $6.2M seed credible despite no track record (tech.az).

What can be applied

Sell the pain users already feel: Nozomio monetized agents losing context with a plug-in that works inside existing tools, proving willingness to pay before the architecture question was settled.

Aftermath

As of 2026-09-02 Nozomio is live with a three-person team in San Francisco, still focused on Nia as an MCP context server for Cursor, Continue and Cline, with the founder saying the team will stay under 20 until IPO. The bet now stands or falls on whether context infrastructure becomes a permanent layer for agents or gets absorbed by model providers and IDEs.

Sources

spotted an error? The archive wants to know.

Your turn

You just read one. Describe what you are building, and see who is betting on the same thing.

Free account · 3 free questions · no card

Related cases