Embedded LLM has introduced TokenVisor Spaces, a platform designed to facilitate governed agentic AI services on AMD-powered infrastructure. The launch coincides with AMD Advancing AI 2026, emphasizing AI infrastructure and enterprise deployment.
TL;DR
- Embedded LLM launches TokenVisor Spaces for AMD-powered AI clouds, enabling governed agentic AI services.
- The platform integrates AMD EPYC CPUs and Instinct GPUs, providing isolated workspaces for agents with policy, metering, and audit capabilities.
- TokenVisor Spaces aims to help AI cloud operators and enterprises deploy production-ready agentic AI services.
What happened
Embedded LLM, an agentic inference infrastructure company and part of the AMD Instinct™ AI & HPC Software Ecosystem, has launched TokenVisor Spaces. This platform is designed for AMD-powered AI clouds and enterprises, aligning with the focus of AMD Advancing AI 2026 on AI infrastructure and enterprise deployment.
TokenVisor Spaces enables AI cloud operators to transform AMD-powered systems into governed agentic AI services. It provides agent execution on AMD EPYC™ x86 CPUs, model inference on AMD Instinct GPUs, policy and metering through TokenVisor, and replayable workspaces for enterprises to audit and operate.
The platform is validated on Red Hat OpenShift, offering isolated agent sandboxes with governed LLM access. OpenShift manages containers and GPUs, while TokenVisor enforces entitlements, routing, budgets, rate limits, metering, and auditability.
Why it matters
TokenVisor Spaces addresses the need for agents to run tools, browsers, code, files, queues, approvals, and enterprise workflows on x86-64 infrastructure. This makes it relevant for AI cloud operators looking to sell governed agent services and enterprises wanting to integrate agents with existing applications.
The platform's ability to provide persistent workspaces, isolated execution, human approval workflows, replayable event history, and governed model access makes it a comprehensive solution for production agentic AI services.
For developers and startups, TokenVisor Spaces offers a way to deploy agentic AI services on controlled infrastructure, while investors may see potential in the growing market for governed AI services.
Key facts
- TokenVisor Spaces integrates AMD EPYC CPUs and Instinct GPUs for agent execution and model inference.
- The platform is validated on Red Hat OpenShift, providing isolated agent sandboxes with governed LLM access.
- Embedded LLM is a partner in the AMD Instinct AI & HPC Software Ecosystem and a Red Hat ecosystem partner.
- TokenVisor Spaces offers persistent workspaces, isolated execution, human approval workflows, replayable event history, and governed model access.
- The platform is available for partner deployment and evaluation by AI cloud operators, private AI environments, and on-premises enterprises.
- Embedded LLM has validated the model-plane data path for KV-cache reuse on AMD Instinct MI355X, demonstrating improved performance metrics.
- TokenVisor Spaces is part of Embedded LLM's suite of products, including vLLM-based serving, TokenVisor for governed model APIs and monetization, JamAI Base for traceable AI operations, and operator-level RL/post-training infrastructure.
Context
The launch of TokenVisor Spaces comes at a time when there is a growing demand for agentic AI services that can integrate with existing enterprise software. The platform's focus on governed, production-ready agentic AI services aligns with the broader industry trend towards deploying AI solutions in controlled environments.
Embedded LLM's collaboration with VAST Data and Tensormesh for platform-scale KV-cache reuse, agent state, trace capture, replay/evaluation, and RL data underscores the importance of efficient context management in long-running AI workflows.
As AI clouds and enterprises look to deploy agentic AI services, platforms like TokenVisor Spaces that provide governed, isolated workspaces and seamless integration with existing infrastructure are likely to gain traction.
