Edge AI in 2026 means hardware that fits on a desk, in a backpack or in a closet, running real AI workloads without the cloud bill. We document what each device can run, from the makers' specs and published measurements, and pair it with the agents that fit.
The pocket AI thesis is simple: hardware in 2026 is good enough that "run your AI on a device you own" is competitive with cloud-hosted alternatives for most personal and small-team use cases. The catch is matching the hardware to the workload.
A Raspberry Pi 5 (8 GB) can host Hermes Agent without browser tools and a Phi-3 mini local model: €100 hardware, ~5W power, an agent that calls Claude or Mistral via OpenRouter and uses local tools. That's the entry point.
A €450 generic Intel mini PC with 32 GB RAM is the workhorse: full Hermes Agent, browser automation, Mistral 7B local fallback, multi-tool workloads without RAM pressure. That's where most pocket AI deployments should land.
A €1,099 Mac Mini M4 with 24 GB unified memory is the local-LLM upgrade: 24.11 tokens per second with a 7B model on llama.cpp's published scoreboard, and near-silent. A €1,899 M4 Pro with 48 GB loads 70B-class models; 64 GB leaves room for the context and the agent.
Above that, you're buying workstation territory and pocket AI stops being the right framing. This hub covers the hardware that fits the pocket AI brief.
Guides
- Pocket AI: which self-hosted agents run on small hardware: Which self-hosted AI agents run on a Raspberry Pi, mini PC, Mac mini, Framework laptop or old phone: an agent-by-device matrix, the limits, six setups.
- Best hardware for local LLMs and AI agents in 2026: Raspberry Pi 5, mini PC, Mac mini or Mac Studio for a local LLM or a self-hosted AI agent: what each budget buys, with makers' specs and published speeds.
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.
- Intel NUC 13 and Geekom IT13: Intel NUC 13 and Geekom IT13: 13th-gen Intel Core mini PCs with 16 to 64 GB of RAM, room for Hermes Agent with browser automation and a 7B model beside it.
- Mac Mini M4 / M4 Pro: Mac mini M4 and M4 Pro for local LLMs: 16 to 64 GB of unified memory at 120 or 273 GB/s, the two specs that decide which models load and how fast.
- Lenovo ThinkCentre M75q (used): Lenovo ThinkCentre M75q, used: a €150-250 Ryzen mini PC with 16 GB of RAM and an NVMe slot, enough for Hermes Agent with the browser tool.
- Framework Laptop 13 / 16: Repairable, modular laptop with strong Linux support. The right choice if you want a portable agent host that doubles as a daily driver.
- Minisforum UM790 Pro: Minisforum UM790 Pro: Ryzen 9 7940HS with Radeon 780M graphics and up to 64 GB of DDR5, for a Linux agent host that also runs local models.
- 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.
Terms
Pocket AI · Edge AI · VPS · Self-hosted AI