F1 · Edge AI & Agents
DGX Persistent Agent: Hermes + Qdrant
Overview
On an NVIDIA DGX Spark, this guide composes Hermes Agent, Ollama, Qdrant and Open WebUI with Docker into a local agent runtime with a persistent workspace, semantic memory, embeddings and contextual retrieval. The Grace CPU coordinates events, memory and service lifecycle, composing model inference, a vector database and the user interface into a full, independently swappable service path. This modular reference architecture gives developers an observable, extensible persistent-agent baseline that concretely shows the CPU's key role in local AI orchestration.
Scope of verification
- It is a persistent-agent reference runtime on a single DGX Spark, validating that Hermes, Ollama, Qdrant and Open WebUI together provide workspace, memory, retrieval and interaction.
- Dependencies include the Grace Blackwell GB10, Docker, matching container images, local models, embedding configuration, Qdrant storage and sufficient GPU/system memory.
- Scope centers on a runnable MVP and its state-persistence approach; it does not cover long-run soak testing, HA, backup/recovery, multi-tenant isolation, secrets management, production security or operations SLOs.
Canonical source: https://learn.arm.com/learning-paths/laptops-and-desktops/dgx_persistent_agent/
This page is a guided summary with boundary notes, not an official reproduction.
This page is a guided summary with boundary notes, not an official reproduction.