Self-hosted AI hosting in 2026 spans free Cloudflare Workers, €5/month VPS, €1,899 Mac Mini and €4,500 Mac Studio. Each has a specific shape of workload it fits. This hub maps which is which.
The free tier first. Cloudflare Workers covers 100K requests per day at zero recurring cost for code that runs as a plain Worker; its runtime rules out browser automation and native binaries. Moltworker is not in that tier: it runs OpenClaw in a Sandbox container that needs the Workers Paid plan plus container time, about $34.50 a month around the clock by its README. Until October 2026 this paragraph put Moltworker in the free tier.
The €5-15/month tier is where most self-hosters should sit. Hetzner CX22 (€5, 2 vCPU, 4 GB RAM) runs Hermes Agent without browser tool comfortably. Add the browser tool and you want CCX13 (€38, dedicated CPU, 16 GB RAM) instead. Hostinger and Contabo offer similar tiers.
The €30-100/month tier is where managed-hosting services compete with VPS, though their plans start lower and go higher. On their own pricing pages, read on 18 October 2026, ClawRift asks $19 or $49 a month (less billed yearly), NitroClaw $20 or $100 with AI credits included, and ClawGo $20, $49 or $200 with model usage billed on top. Until October 2026 this paragraph also named DeployHermes, whose site now presents Hermes Agent and sells no plan. For anyone who wants the agent without the sysadmin work, these are reasonable. For anyone who values the control of self-hosting, they're not.
Above that, you're in workstation territory. Mac Mini M4 (€1,099-1,899) and Mac Studio M3 Ultra (€4,500+) are the small-form-factor leaders for local-LLM-heavy workloads. Custom workstations with discrete GPUs go further but trade the unified-memory advantage.
Pick the tier that matches your workload, not the tier that matches your aspirations.
What this page says about each agent's defaults was read in its own documentation on 2026-10-18: Moltworker README (main, 7b00c1d): Workers Paid plan, container costs.
Guides
- 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.
- 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.
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.
- Moltworker: Cloudflare's experimental package that runs OpenClaw inside a Cloudflare Sandbox container behind a Worker. Needs the Workers Paid plan.
- ZeroClaw: Single Rust binary from ZeroClaw Labs, MIT or Apache-2.0. Works with Anthropic, OpenAI, Ollama and about twenty other providers, and turns on an OS sandbox it detects.
- ClawRift: Managed OpenClaw hosting, live on Telegram (Discord and WhatsApp "Soon"). $19 or $49 a month, $14 or $37 billed yearly. Your own model API key.
- NitroClaw: Managed assistant built on OpenClaw, in Telegram and Discord: $20 a month self-serve, or $100 with setup and a monthly call. AI credits included.
- ClawGo: Managed OpenClaw hosting for Telegram, Discord and WhatsApp. No free plan: $20, $49 or $200 a month, with model usage bought separately as credits.
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.
- Mac Studio M3 Ultra: Apple's 2025 local-LLM workstation: M3 Ultra, 96 GB of unified memory configurable to 256 GB, 819 GB/s. Since replaced by the M5 Ultra model.
- 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.
Providers
- Hetzner: German hosting company: cloud servers and dedicated servers. PocketClaw itself runs on Hetzner servers in Germany.
- Cloudflare Workers: Serverless platform with a free plan of 100,000 requests a day. Moltworker, which runs OpenClaw in a Sandbox container behind a Worker, needs the paid plan.
- Tailscale: Mesh VPN built on WireGuard: a common way to keep a self-hosted dashboard off the public internet.
Terms
VPS · Hetzner · Cloudflare Workers · Tailscale · Caddy · Docker Compose