F2 · Edge AI & Agents
OpenClaw Continuum: Local-First Assistant
Overview
On an NVIDIA DGX Spark and a Radxa Orion O6, this guide deploys the OpenClaw local-first assistant, integrating Telegram, local LLM inference, Ollama embeddings, Qdrant persistent memory, document RAG, browser search and scheduled tasks. The workflow connects vLLM on DGX Spark and llama.cpp on the Armv9 CPU-only platform through an OpenAI-compatible API, validating that the same application flow runs across different Arm compute configurations. The architecture decouples the inference backend from the agent workflow and clearly shows the data boundary among CPU orchestration, local data processing and external services.
Scope of verification
- It is a single-user local-first assistant reference runtime, validating the integration of OpenClaw, memory, RAG, tools, scheduling and two Arm inference backends.
- Dependencies include a DGX Spark or Radxa Orion O6, OpenClaw, Telegram, Qdrant, Ollama, vLLM or llama.cpp, local models, a document corpus and external search/weather services.
- Local-first is not fully offline: Telegram, model downloads, browser search and external queries cross the local boundary; scope also excludes multi-model routing, multi-agent handoff, cross-hardware performance comparison, HA, multi-tenant isolation and production security.
Canonical source: https://learn.arm.com/learning-paths/laptops-and-desktops/openclaw_continuum/
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.