AMD Releases Loom SDK for Agentic AI Workloads
The initial v0.1 release adds a new compiler stack to the ROCm ecosystem to support execution across CPUs, GPUs, and NPUs.
Updated on Oct. 29, 2026 in Semiconductors
Live Poll
Do you trust new open-source development tools to effectively improve the software you use?
AMD released version 0.1 of its ROCm HRX Loom SDK, a developer toolset designed to optimize agentic workloads across hardware platforms. The SDK includes a compiler, runtime, Python APIs, and command-line tools to support execution on CPUs, AMD GPUs, and AMD NPUs.
Why it matters
Loom aims to streamline the development of agentic AI workloads, which require efficient task execution across heterogeneous hardware architectures. This release marks an initial step in unifying deployment targets ranging from data center GPUs to local client-side CPUs.
The Loom SDK v0.1 provides a compiler and IR stack with support for AMDGPU HSA emission, SPIR-V/Vulkan, x86, and WebAssembly. These components facilitate cross-platform execution compared to traditional vendor-specific proprietary toolkits.
The players
AMD
A semiconductor company that designs CPUs, GPUs, and NPUs for data center, AI, and consumer computing markets.
The details
The Loom SDK functions by separating model management from deployment via specialized tools. Developers use the 'loom-fetch' utility to install model programs and assets, while the 'loom-serve' utility handles loading the accelerator card and managing inference requests over HTTP. By providing a common intermediate representation (IR) stack, the framework aims to simplify how agentic AI applications interface with AMD’s CPU, GPU, and NPU hardware.
Timeline
The Loom SDK v0.1 release became available on October 29, 2026.
The Tech Race
This release positions AMD deeper into the agentic AI market, competing against broader industry efforts to standardize cross-platform AI deployment. It extends the existing ROCm software platform to address specific hardware-level bottlenecks encountered when running autonomous agent workflows.
Developers can currently access the SDK via the hrx-loom package on Python pip or through GitHub for integration into existing AI pipelines. The tools enable immediate testing on AMD hardware, though developers should expect evolving support for x86 and WebAssembly targets.
The takeaway
AMD is targeting the architectural demands of agentic AI with a new cross-platform compiler stack. Future updates to track include performance benchmarks for x86 and WebAssembly compilation compared to existing native toolchains.
Further reading
For more on evolving processor architectures and toolsets, visit our Semiconductors section.
More information
Access the documentation and build tools at the Loom SDK GitHub repository.
Source note: This article includes information reported by Phoronix.
Live Poll
Do you trust new open-source development tools to effectively improve the software you use?
