Nozomio builds context APIs for coding agents — 25% of current YC batch paying, $11.5K MRR

Summary

Arlan Rakhmetzhanov, 17, is building Nozomio, an API that gives coding agents persistent, clean context by letting developers index documentation or codebases directly into tools like Cursor.

The problem is one most Cursor users have hit: copying and pasting docs into the chat window pollutes the context window and disappears when the session resets. Research from Chroma shows model performance degrades sharply past roughly 4,000 tokens. On the architecture question of RAG versus expanding context windows, Arlan notes that Cursor uses both local indexing and a vector store, while Claude Code skips indexing entirely and relies on agentic grep. RAG remains effective when combined with graph RAG approaches, though the industry hasn't resolved the debate.

Traction

Within three weeks of outreach, Nozomio has 25% of the current YC batch paying, plus 5% of the prior spring batch. That totals more than 90 companies and $11.5K MRR. The business is already profitable, with deal revenue covering all infrastructure costs.

Arlan raised $1 million in a couple of days in London from LocalGlobe before applying to YC. He got in on his third attempt. His father had applied three years earlier and been rejected, which motivated him to keep reapplying. He closed a further fundraise a few weeks before Demo Day.

At 15, he built a database connecting high school students from underrepresented backgrounds to US summer programs, scaled it to 20,000 active users, and made roughly $2,000.

With profitable unit economics, 90-plus paying companies, and a founder who moved from London dropout to YC in under a year, Nozomio signals that context management for coding agents has genuine willingness to pay before frontier model providers have settled on a standard architecture.