Raspberry Pi for self-hosted AI

What a Pi 5 actually does for self-hosted AI agents: the workloads that fit, the ones that don't, and the budget combos.

The Raspberry Pi 5 is the default starting hardware for self-hosted AI in 2026. It's also the most over-promised: the Pi can do real things, but not everything you'll read on YouTube.

What works on a Raspberry Pi 5 (8 GB) in 2026: Hermes Agent without the browser tool, Nanobot, OpenClaw 2026.4 with browser disabled. Power draw stays under 12W typical; the official active cooler keeps thermals in line for sustained workloads. Add the M.2 NVMe HAT (€60) for any deployment that runs longer than a week: microSD lifespan under sustained agent writes is months, not years.

What does not work: Llama 3 8B at any usable speed (forget it on the Pi), browser automation at scale (Chromium is too heavy), multi-user workloads.

The Orange Pi 5 Plus is the credible alternative: up to 16 GB of RAM, RK3588 with NPU, M.2 NVMe out of the box. It's meaningfully more capable on paper. The catch is software ecosystem: Pi Foundation has the largest hobbyist community in computing, Orange Pi has Armbian. If you'll value polish over capability, stay on the Pi 5. If you'll value capability and you're comfortable on Armbian, the Orange Pi 5 Plus is better.

The Pi Zero 2 W (€20) is too small to host a primary agent but excellent as an edge MCP tool host in a distributed setup. Pair a primary Pi 5 with 3-5 Pi Zeros and you have a multi-tool agent network for under €200.

Guides

Agents on this topic

  • Hermes Agent: Agent from Nous Research (MIT, Python). Runs shell commands on the host by default, with an approval step for dangerous ones; Docker and cloud sandboxes are options. Works with many LLM providers.
  • Nanobot: Personal AI agent in Python (MIT) with a browser UI and chat apps. Native Anthropic, OpenAI and OpenRouter providers plus local servers; its shell sandbox is off until you switch it on.
  • Moltworker: Cloudflare's experimental package that runs OpenClaw inside a Cloudflare Sandbox container behind a Worker. Needs the Workers Paid plan.

Hardware on this topic

  • Raspberry Pi 5: Raspberry Pi 5 for self-hosted AI: four Cortex-A76 cores, 1 to 16 GB of LPDDR4X. Hosts Hermes Agent (no browser tool) or Nanobot; local models up to about 3B.
  • Raspberry Pi Zero 2 W: €20 SBC with 512 MB RAM. Useful as an edge agent endpoint (sensor reader, tool host) but not a primary agent host.
  • Khadas Edge2 Pro: The Khadas Edge2 in its Pro configuration: Rockchip RK3588S2, 16 GB of LPDDR4X and 64 GB of eMMC on a small board. Polished hardware, smaller community.

Terms

Pocket AI · Edge AI · Phi-3 · Ollama · MCP server