Skip to main content
used mini pcs

Agents that never sleep

The used mini PC we'd buy to run AI agents 24/7

Your AI agents don't need a GPU at home. The model answers from the cloud, and the box in your closet runs the agent harness, its tools and its browsers around the clock. We'd use a used business mini PC for that. It idles at around the wattage of a light bulb, takes 32 to 64 GB of RAM, and costs nothing per month once you own it.

What you need

SpecRecommendedWhy
CPU generationIntel 8th gen or newer, or Ryzen 4000 (Renoir) or newer; 6 cores or more preferredThe agent loop itself is light. What agents do with it, like installing packages, running test suites, building code and driving a browser, is bursty CPU work. A 6-core Coffee Lake or Ryzen PRO GE chip lets two or three agents run in parallel without queuing. Claude Code needs an x64 or ARM64 processor, and every catalog model has one.
RAM16 GB minimum, 32 GB comfortable, 64 GB for many parallel or browser-heavy agentsClaude Code lists 4 GB+ RAM as its floor. Anthropic's Agent SDK hosting guide suggests 1 GiB of RAM per agent as a starting point and warns that memory grows with session length and tool use. Headless browsers, Docker containers and language servers stack on top of that.
Storage typeNVMe SSD, 500 GB or largerThe Agent SDK guide suggests 5 GiB of disk per agent as a baseline. Repos, node_modules folders, browser binaries, Docker images and session transcripts add up quickly, and NVMe keeps installs and git operations fast.
NICsOne wired 1 GbE portAgent traffic is mostly outbound HTTPS to the model API plus the odd webhook, so bandwidth won't hold you back. A wired connection is simply more reliable than Wi-Fi for a box that has to stay reachable.
Idle wattsAbout 10 to 15 W at the wall for a 1-liter business tinyAn agent box spends most of its life waiting for the next job, so idle draw is what you pay for. A ThinkCentre M720q measured 11 to 14 W idle, and ServeTheHome put its whole first batch of tiny/mini/micro test units at about 10 W.
NoiseFan-cooled 1-liter tiny is fine; near-silent at idleThe M720q's idle noise is insignificant, and the M920q only becomes audible under sustained CPU load. Agent work comes in bursts, so the fan spends most of its time near idle.
Remote managementNice to have: Intel vPro (AMT) or AMD DASHOn a headless box, out-of-band management lets you power-cycle it or reach the BIOS without carrying a monitor over. vPro M920q configurations have Intel AMT 12.0, and M75q Gen 2 units with the Realtek RTL8111FP NIC have DASH 1.2.

Which mini PCs fit

A model gets the “AI agent server” badge when it:

  • Intel 8th gen or newer, Intel N-series, or AMD Ryzen 2000 or newer (not Ryzen Embedded)
  • Takes at least 16 GB of RAM (manufacturer's stated maximum)
  • Has an M.2 slot that takes an NVMe SSD
  • Has wired Ethernet

We only use the makers’ specs. 36 of the 58 models we cover qualify.

Best overall

Lenovo ThinkCentre M75q Gen 2 Tiny

Several agents running headless browsers side by side need RAM, and this one takes up to 64 GB of DDR4-3200 in two SO-DIMM slots. Its Ryzen PRO GE chips go up to 8 cores and 16 threads (Ryzen 7 PRO 4750GE/5750GE). There's an M.2 NVMe slot plus a 2.5-inch bay, and units with the RTL8111FP NIC get DASH remote management, all in a quiet 1-liter box.

On eBay now: Linux Mint is already on it, so it's a ready Home Assistant or Docker box at $200 delivered. Ask what size the drive is. $200.00

View on eBay ↗: Lenovo ThinkCentre M75q Gen 2 Tiny Ryzen 3 Pro 4350GE | 16GB RAM | Linux Mint OS

Best value

Lenovo ThinkCentre M720q Tiny

It's one of the most common off-lease tinies, so you can be picky about the spec. Lenovo built it with 8th/9th-gen T-series CPUs from Celeron up to the 8-core i7-9700T and i9-9900T, but the six-core i5-8500T and i7-8700T are what you'll mostly see. It has two DDR4-2666 SO-DIMM slots (32 GB official), an NVMe slot plus a 2.5-inch bay and an Intel I219-V NIC, and it idles at 11 to 14 W. A six-core unit with 32 GB handles a few agents comfortably.

On eBay now: A quad-core i3-8100T with an M.2 SSD for $115 delivered. Fine for Home Assistant or Pi-hole, but ask about the adapter. $115.00

View on eBay ↗: Lenovo ThinkCentre M720q Intel Core i3 8100t 8GB 256GB SSD M2

What it costs to leave on all year

An always-on box spends almost all its time near idle, so idle draw sets the bill. Here’s the sum.

idle watts × 24 h × 365 days ÷ 1,000 = kWh per year · kWh × $0.1831/kWh = dollars per year

MachineIdlekWh / yearCost / year
1-liter business tiny (Lenovo ThinkCentre M720q, Pentium Gold G5400T), idle at the wall, measured by ServeTheHome, low end source11 W96.4$17.64
1-liter business tiny (Lenovo ThinkCentre M720q, Pentium Gold G5400T), idle at the wall, measured by ServeTheHome, high end source14 W122.6$22.46
Beelink SER5 (Ryzen 5 5500U / 5600H / Ryzen 7 5700U) (measured, 5.3 W upper figure) source5.3 W46.4$8.50
Minisforum UM790 Pro (Ryzen 9 7940HS) (measured, 7 W upper figure) source7 W61.3$11.23
Intel NUC 13 Pro (NUC13ANK / NUC13ANH) (measured, 8 W upper figure) source8 W70.1$12.83
Beelink SER6 Pro (Ryzen 7 6800H / 7735HS) (measured, 8.2 W upper figure) source8.2 W71.8$13.15
ASUS NUC 14 Pro (measured, 9 W upper figure) source9 W78.8$14.44
Beelink GTR6 (Ryzen 9 6900HX) (measured, 10 W upper figure) source10 W87.6$16.04
Dell OptiPlex 3070 Micro (measured, 10 W upper figure) source10 W87.6$16.04
Dell OptiPlex 7080 Micro (measured, 11 W upper figure) source11 W96.4$17.64

Electricity at 18.31¢/kWh, the U.S. average residential price, July 2026 (EIA). Your own rate is on your bill, so swap it into the sum. Real draw rises above idle whenever the box is working, and it depends on the drives, RAM, BIOS power settings and OS.

What the box does, and what it doesn't

Get this split clear before you buy anything. When you run Claude Code or an app built on the Claude Agent SDK, the model runs in Anthropic's cloud and you reach it over the API. Claude Code's own requirements list an internet connection as mandatory, along with a Pro, Max, Team, Enterprise or Console account. The mini PC runs everything around the model. That's the agent loop, the shell it runs commands in, git, your test suites, MCP servers, headless browsers, and the scheduler that kicks jobs off at 3 a.m. So a modest used box does the job, and you don't need a GPU for it. Running the model itself on your own hardware is a different and much heavier workload, covered in running local LLMs on a mini PC.

Claude Code and Agent SDK requirements

Claude Code's published system requirements are modest. It runs on macOS 13+, Windows 10 1809+ or Windows Server 2019+, Ubuntu 20.04+, Debian 10+ or Alpine 3.19+, with 4 GB+ RAM and an x64 or ARM64 processor. The native installer doesn't need Node.js at all. Only the optional npm package asks for Node.js 22 or later. If you build your own agent with the Agent SDK, the runtime is Python 3.10+ or Node.js 18+, and the SDK bundles its own Claude Code binary.

The SDK docs earn their keep on sizing. Each agent session is a separate long-lived subprocess with its own shell, working directory and transcript on disk. Anthropic suggests 1 GiB RAM, 5 GiB disk and 1 CPU per agent as a starting point, and says memory grows with session length and tool activity. It also gives a simple formula, where agents per host equals host RAM minus overhead, divided by the peak RAM one session actually reaches. Measure a real session with your real tools instead of trusting the floor. In practice, 32 GB is a comfortable target for a few agents at once, and 64 GB is where browser-heavy fleets stop fighting each other.

Scheduled agents with cron or systemd timers

For unattended jobs, start with Claude Code's non-interactive mode. `claude -p "…"` runs one task and exits with code 0 on success and non-zero on failure, so cron or a systemd timer can branch on the result. The docs recommend `--bare` for scripted runs. It skips auto-discovered hooks, plugins and MCP servers, and it needs `ANTHROPIC_API_KEY` set. `--output-format json` returns the result plus an estimated cost figure, and `--permission-prompts none` denies anything that would otherwise wait for a human. Add a tight `--allowedTools` list and a dedicated Linux user, and a nightly dependency-update or report-writing agent takes a few lines of config.

Browser agents are the heavy part

Agents that drive a real browser through Playwright are the hungriest thing you'll put on this box, because every session is a full Chromium instance. Playwright currently requires Node.js 22, 24 or 26, and on Linux it lists Debian 12/13 and Ubuntu 22.04, 24.04 and 26.04 on x86-64 or arm64. Playwright doesn't publish a RAM figure per browser, so measure. Launch your usual number of sessions, watch peak memory, and size up from there. That's the main reason the picks on this page lean toward 64 GB-capable machines. Ubuntu 24.04 LTS is on both Claude Code's and Playwright's support lists, which makes it the easy OS choice.

MCP servers and webhooks

MCP servers come in two standard transports. With stdio, the agent launches the server as a local subprocess, which costs a little RAM per server and nothing else. With Streamable HTTP, the server is an independent process with its own endpoint. The spec says local servers should bind to 127.0.0.1 instead of every interface and should authenticate connections. Keep them on localhost and you avoid most of the risk.

Webhooks, like a GitHub push, a form submission or a payment event, need a way in from the internet. Use a tunnel instead of forwarding a port on your router. Cloudflare Tunnel's `cloudflared` makes an outbound-only connection, so the box needs no publicly routable IP and you can block all inbound traffic. Tailscale Funnel routes public traffic to a local service on ports 443, 8443 or 10000, TLS only. It's available on all Tailscale plans and has bandwidth limits you can't configure, which is fine for webhooks.

Why a mini PC beats a laptop or a VPS for this

  • An old laptop wants to sleep when the lid closes, rides on Wi-Fi, and keeps its battery at full charge for months. A mini PC has none of those habits, takes a wired connection, and can be set in BIOS to power back on after an outage.
  • A cloud VPS bills you every month, forever. A mini PC is a one-time purchase plus electricity, and your repos, browser profiles, API keys and the files agents work on stay in your house instead of on someone else's disk.
  • A VPS still wins on redundant power, a static IP and nobody tripping over the cable, because it lives in a data center. If your agents serve paying customers, host the public endpoint in the cloud and let the home box do the private work.

Power is the running cost that matters. At the 11 to 14 W a ThinkCentre M720q draws at idle, the yearly electricity estimate on this page stays small, and agent bursts barely move the average. Anthropic's own hosting guide points out that model token spend typically dwarfs infrastructure cost, so the box will rarely be your biggest line item.

How much machine you need

Choosing between the three big business brands? Our ThinkCentre vs OptiPlex vs EliteDesk comparison covers how they differ on RAM limits, storage layout and remote management.

Setup outline

  1. 1Ask the seller before you pay whether a BIOS supervisor (administrator) password is set. It locks boot order, UEFI and Secure Boot settings, which a headless agent box needs you to change, so get it cleared or pass.
  2. 2On first boot, open the BIOS security page and find the Computrace or Absolute setting. Absolute's persistence module sits in the firmware of many makers' machines and can put its agent back after a wipe. If it reads activated, ask the seller for proof the last owner released the device.
  3. 3Flash the newest BIOS from the maker's support site, found by serial number. On off-lease units, old firmware is behind most odd memory and NVMe behavior.
  4. 4Install RAM as a matched pair so it runs dual-channel, and fit an NVMe SSD of 500 GB or more.
  5. 5Run a full pass of MemTest86+ from a USB stick before you install anything. When agents crash at random on a used box, bad memory is the usual cause.
  6. 6In BIOS, set the machine to power on after AC loss so it comes back by itself after an outage, then install Ubuntu Server 24.04 LTS.
  7. 7Create a dedicated non-admin user for agents, install Claude Code with the native installer, and run `claude doctor` to confirm the install.
  8. 8If you use browser agents, install Node.js 22 or later and Playwright with its Chromium build and system dependencies.
  9. 9Schedule jobs with systemd timers or cron calling `claude -p` with `--bare`, a narrow `--allowedTools` list and `--permission-prompts none`. Expose webhooks through Cloudflare Tunnel or Tailscale Funnel instead of opening router ports.

AI agent server: common questions

Does the mini PC run the AI model?

Not for Claude Code or Agent SDK agents. The model runs in Anthropic's cloud and you reach it over the API. The mini PC runs the agent harness, its shell, tools, MCP servers and browsers, which is why a used 1-liter business PC is enough. Running the model locally is a different job, covered in local LLMs on a mini PC.

How much RAM do I need to run Claude Code 24/7?

Claude Code itself lists 4 GB+ RAM. For an always-on server, plan on 16 GB for one or two agents without a browser, 32 GB for several agents or Playwright, and 64 GB for many browser sessions at once. Anthropic's Agent SDK guide suggests 1 GiB per agent as a floor and recommends measuring the peak memory of a real session.

Is a mini PC better than a VPS for AI agents?

For private agents, usually yes. There's no monthly fee, you get plenty of RAM in a one-time purchase, and your files and credentials stay at home. Pick a VPS when you need a static public IP, data-center uptime, or a service customers depend on.

Can an Intel N100 mini PC run AI agents?

It can run one or two light scheduled agents. Most N-series machines are limited to 16 GB on a single memory channel, and that gets tight once headless browsers or Docker are involved. We'd buy a used 6-core business tiny with 32 GB instead. See N100 vs N150 vs N305.

How do I receive webhooks without opening ports?

Use an outbound tunnel. Cloudflare Tunnel's cloudflared connects out to Cloudflare, so you can block all inbound traffic. Tailscale Funnel publishes a local service on port 443, 8443 or 10000 over TLS.

Is it safe to leave an AI agent running unattended?

It is if you scope it tightly. Run it as its own user and list only the tools it needs with --allowedTools. Use --permission-prompts none so anything outside that list is denied instead of waiting. Keep MCP servers bound to localhost, and never give the agent's account admin rights.

Sources

Run more on the same box

Upgrading? Sell the old one to us.

Tell us what you have (model, CPU, RAM, drive, whether there’s a BIOS password), add a few photos and your price, and we’ll reply with an offer by email. No listing, no fees.

Get an offer