# ThoughtForge AI

> ThoughtForge AI is a Berkeley-based robotics software company building what it calls Universal Robotic Intelligence: an Active Inference control platform that lets existing industrial robots, drones, and crawlers adapt in real time to messy, unpredictable environments. Its lightweight models run on a single CPU at the edge, claim over 99% accuracy on tasks as designed, and reportedly require roughly 90,000x less training data than conventional deep learning. Founded by veterans of Microsoft's Bonsai AI team, ThoughtForge sells to manufacturing, energy, aerospace, and logistics operators who want to upgrade the robots they already own rather than replace them.

- **Founded:** 2020
- **Headquarters:** Berkeley, California, United States
- **Founders:** Matthew Brown (Founder & CEO), Harneet Singh (Co-Founder & Chief Product Officer)
- **Team size:** ~11 employees
- **Products:** Replicate, Control, See
- **Notable:** Team pedigree from Bonsai AI, which built the world's first production-grade deep reinforcement learning platform (acquired by Microsoft in 2018)., Advised by Prof. Karl Friston, originator of the Free Energy Principle and Active Inference., Claims over 99% accuracy on tasks as designed while using roughly 90,000x less data than conventional ML.

## Products & services

- **Replicate** — Converts existing AI models into ThoughtForge's Active Inference format to improve accuracy, cut compute requirements, and simplify deployment.
- **Control** — Universal robotic control that lets robots operate in dynamic, unpredictable environments, adapting in real time on standard CPUs at the edge.
- **See** — Unified fleet monitoring and performance visibility across a robot fleet through a single API.

## Achievements

- Team pedigree from Bonsai AI, which built the world's first production-grade deep reinforcement learning platform (acquired by Microsoft in 2018).
- Advised by Prof. Karl Friston, originator of the Free Energy Principle and Active Inference.
- Claims over 99% accuracy on tasks as designed while using roughly 90,000x less data than conventional ML.
- Reported compatibility with 78% of existing industrial robots, running on a single CPU at the edge without WiFi.
- Selected as an ARM Institute member and project awardee for deformable-part insertion.

## Latest updates

- **2024-02** — Website and product suite (Replicate, Control, See) publicly positioned around Universal Robotic Intelligence and OEM compatibility.
- **2024** — ARM Institute 'Five Questions' feature and funded project on soft-part insertion for automotive, aerospace, and electronics manufacturing.
- **2024** — Industry Today feature on breaking robotic AI limits with real-time adaptation and Adaptive Specialized Intelligence (ASI).

## Links

- Website: https://thoughtforge.ai
- LinkedIn: https://www.linkedin.com/company/thoughtforge

---

Profile page: https://yespress.io/thoughtforge-ai
Published by YesPress — https://yespress.io
Last updated: 2026-07-02