# HUD

> HUD is a San Francisco startup (Y Combinator W25) that builds infrastructure for evaluating and training AI agents. Its platform lets teams wrap real software - browsers, spreadsheets, terminals, apps - into reinforcement-learning environments and evaluations, then run agents against them at scale to measure whether they actually work. HUD ships an open-source Python SDK plus a cloud platform and a vendor marketplace that connects environment builders with frontier AI labs.

- **Founded:** 2025
- **Headquarters:** San Francisco, United States
- **Founders:** Jay Ram (Co-CEO / CEO), Lorenss Martinsons (Co-CEO / Chief Product Officer), Parth Patel (CTO)
- **Team size:** ~10-15 employees
- **Products:** Environment SDK (hud-python), Training + Eval Platform, HUD Vendor Marketplace, inference.hud.ai
- **Notable:** Accepted into Y Combinator's Winter 2025 batch, Reported 1.3M+ task runs, 2,500+ environments and 15M+ inference calls on the platform, Over 50 businesses building RL environments on HUD

## Products & services

- **Environment SDK (hud-python)** — Open-source (MIT) Python SDK to define evaluations, environments, verifiers and rewards, wrapping real software as agent-callable tools in isolated Docker containers. Protocol-first design across coding, browser, computer-use and robotics.
- **Training + Eval Platform** — Cloud platform to train and evaluate models on environments at scale - thousands of concurrent instances, agent trace investigation, live telemetry, automated QA auditing (failure analysis, reward-hacking detection, false negative/positive detection).
- **HUD Vendor Marketplace** — A marketplace connecting post-training environment suppliers ('data entrepreneurs') with AI labs that want to buy or commission RL environments and evals.
- **inference.hud.ai** — OpenAI-compatible model inference endpoint used to run agents against environments.

## Achievements

- Accepted into Y Combinator's Winter 2025 batch
- Reported 1.3M+ task runs, 2,500+ environments and 15M+ inference calls on the platform
- Over 50 businesses building RL environments on HUD
- Co-developed benchmarks including SheetBench-50 and contributed to OSWorld-Verified
- Open-source SDK (hud-python) released under MIT license

## Latest updates

- **2025-07** — HUD comes out of stealth, debuting an all-in-one platform for evaluations on computer-use and browser-use AI agents.
- **2025-03** — Seed round reported; company operating out of San Francisco as a YC W25 company.
- **2025** — Launched the HUD Vendor marketplace connecting environment builders with AI labs, and expanded the open-source hud-python SDK across coding, browser, computer-use and robotics.

## Links

- Website: https://www.hud.ai
- LinkedIn: https://www.linkedin.com/company/hud-evals
- Twitter/X: https://x.com/hud_evals
- GitHub: https://github.com/hud-evals

---

Profile page: https://yespress.io/hud
Published by YesPress — https://yespress.io
Last updated: 2026-07-31