Helicone
co-founder & CEO / YC W23 / 2023–2026
The bet
Helicone is an open-source LLM observability platform. I co-founded it and ran it as CEO out of Y Combinator's W23 batch, launching in April 2023.
The thesis was simple. Every hype cycle produces a massive observability market. Web had it, mobile had it, cloud had it, and AI was going to be bigger than all of them. Someone needed to build the platform that ingests LLM traffic and makes sense of it, and I wanted that to be us. When a company ships an AI product, the first question after launch is always the same: what are users actually doing with this, what is it costing us, and where is it failing? Helicone answered those questions with one line of integration.
Being open source was part of the bet too. Observability sits in the request path, and companies are rightly picky about who sees their traffic. Letting anyone read the code or self-host it made the trust conversation short.
Timeline
The Product Hunt run is worth a note. We automated most of the launch itself and wrote up how. Finishing #1 Product of the Day felt appropriate for a company whose whole product is watching software do work.
Numbers
How it worked
Under the hood, almost everything ran through one shared ClickHouse table called request_response_rmt. Dashboards, aggregations, alerts, exports: one table. We even exposed customer-facing SQL on top of it, which sounds reckless until you enforce isolation with row-level security. I wrote about how that works in this post. Keeping the data model that boring is a big part of why a small team could process trillions of tokens without the architecture falling over.
What happened next
In March 2026, Helicone joined Mintlify to build the context layer for AI agents. Observability taught us what agents do with information. Mintlify is where we get to shape the information they consume. I shared the announcement when it happened.