Gimlet Labs

What Gimlet Labs does

Gimlet Labs is an applied research lab dedicated to building next-generation AI infrastructure. They focus on transforming computing systems to efficiently and scalably serve AI workloads. Their work includes serverless inference for AI agents, autonomous kernel generation, and research into AI datacenter scheduling, hybrid edge/cloud partitioning, universal AI compilers, and headless hardware architectures for AI inference.

Products

• Gimlet cloudProvides serverless inference for AI agents, supporting simple to complex multi-agent systems with custom logic and data sources. It handles scheduling, orchestration, and optimization, allowing users to import existing agentic pipelines and chain multiple models with non-model stages and custom data sources.
• kforgeAutonomously generates optimized low-level kernels directly from PyTorch. It uses a multi-agent system with shared memory to explore designs, enforce correctness, and identify the fastest kernels, accelerating training and inference across CUDA, ROCm, and Metal backends without manual kernel writing.

Key capabilities

• Serverless inference for AI agents
• Support for complex multi-agent systems
• Custom logic and data source integration
• Seamless scaling for AI agents
• Automated scheduling, orchestration, and optimization
• Autonomous kernel generation from PyTorch
• Performance acceleration for training and inference workloads
• Support for CUDA, ROCm, and Metal backends
• AI agent architectures for automated kernel generation
• SLA-aware dynamic datacenter scheduling of AI agent workloads
• Hybrid edge/cloud workload partitioning and orchestration
• MLIR-based universal AI compiler for heterogeneous hardware
• Headless hardware architectures for serving AI inference
• Cost-aware optimization frameworks for AI workloads

Use cases

• Running AI agents from simple to complex
• Importing and chaining existing agentic pipelines
• Accelerating AI training workloads
• Accelerating AI inference workloads
• Autoporting AI workloads to new devices
• Optimizing AI workloads for diverse hardware
• Scheduling and partitioning AI agents across distributed hardware
• Improving user experience and reducing provider costs for AI applications
• Running AI workloads on various target systems
• Cost-effective AI inference serving
• Allocating diverse AI tasks at scale in datacenters

AI approach

Gimlet Labs is an applied research lab focused on building next-generation AI infrastructure. They develop platforms and tools for efficient and scalable AI workloads, including serverless inference for AI agents and autonomous kernel generation. Their research areas include AI agent architectures for tuned kernels, dynamic datacenter scheduling, hybrid edge/cloud workload partitioning, universal AI compilers, headless hardware architectures, and cost-aware optimization frameworks.

Tech named: PyTorch, CUDA, ROCm, Metal, MLIR

What it says sets it apart

• Applied research lab focus on next-generation AI infrastructure
• Proprietary serverless inference platform for AI agents
• Autonomous kernel generation technology (kforge)
• Expertise in AI research, production-grade systems, and building successful startups
• Innovative multi-agent system with shared memory for kernel optimization
• Research into novel AI datacenter scheduling and hardware architectures

Funding rounds we track

From the AI funding tracker — rounds as reported by the linked publications.