An observability company
for the agent era.

How we got here

In late 2022, after the public release of GPT-3.5, half a dozen engineers we knew started shipping LLM features into real production systems — Cursor, Notion, Replicate, a stealth analytics startup that's since become a Series B. Within six months, every one of them was running into the same problem: their existing observability stack was the wrong shape.

Datadog could trace the HTTP request that called the LLM. Sentry could catch the exception when the JSON parser blew up. Their internal eval framework caught regressions in CI, sometimes. None of them could answer: what did the agent actually do, why did it cost $7, and is the answer right?

The LLM call graph deserves the same treatment as the HTTP one. We're a year into building it.

So in spring 2023, three of us — Marc Achterberg, Yvonne Chen, and Aiden Park — left Datadog and Honeycomb to start Your Business. We spent the first six months in a basement office in San Francisco's Mission, talking to fifty teams about what they actually needed. The first version of the SDK shipped in October 2023. We hit 100 paying customers in March 2024 and 4,000 in March 2026.

The team today

Your Business is 38 people across two offices — San Francisco and Berlin — and a roster of remote engineers who keep ingest healthy at 14 billion spans a month. Engineering is two-thirds of the company. The rest is design, product, customer success, and a small but mighty go-to-market team led by people we wanted to work with for years.

We've raised $42M total across two rounds led by Index Ventures, with participation from Jeff Dean, Dustin Moskovitz, and a handful of senior engineers we admire. We expect this to be the last round of capital we need.

What we believe

The LLM call graph is the new HTTP request. Observability for it should be a category, not a feature inside another tool. The pricing should match the cost shape (ingest), not the headcount shape (seats). The dashboards should be useful in the first hour, not the second week. And the bar for what "production-grade" means in this space should be set by people who shipped the previous generation of observability tooling — because we know exactly what's hard.