Gimlet Labs

Gimlet Labs builds infrastructure for running AI agents across different types of computing hardware. Gimlet Cloud provides serverless inference, letting developers combine models, search, custom logic, and data sources while the platform handles scheduling and orchestration. Its software splits workloads into stages and maps them to suitable accelerators; customers can use Gimlet-managed infrastructure or deploy the software in their own datacenters. Its kforge toolkit generates optimized kernels from PyTorch using collaborating agents, supporting CUDA, ROCm, and Metal for training and inference.

Founded in 2023, Gimlet’s co-founders are CEO Zain Asgar, Michelle Nguyen, Omid Azizi, Natalie Serrino, and James Bartlett. Its engineering combines workload orchestration, a hardware-agnostic compiler, and autonomous kernel generation to run workloads across heterogeneous systems without requiring developers to rewrite them. The lab also researches kernel verification: an early prototype compares PyTorch references with optimized Triton implementations through symbolic reasoning, finding errors missed by numerical tests. This work complements testing rather than providing a general correctness guarantee.

By March 2026, the company reported that its customer base had tripled since emerging from stealth five months earlier. That month, Gimlet raised an $80 million Series A led by Menlo Ventures, bringing total funding to $92 million to expand its team and inference cloud.