OVHcloud
Self-hosting AI agents (OpenClaw, Hermes Agent, n8n) on low-cost VPS with an EU-hosted model API
We may earn commissions when you shop through the links below. Learn more
OpenClaw, formerly Clawdbot and Moltbot, runs as an always-on gateway. Compare providers with official templates or documentation for hosting it, and check what the docs say about sizing and security.
“Hetzner also runs a free hosted OpenClaw private beta on its Experiments platform: experimental, no SLA, no backups, limited to users with a linked Hetzner Forum account, and announced to become paid after the experimentation phase.”Hetzner provider profile
Model API for OpenClaw · Documented setup
Modal provider profileModel API for OpenClaw · Documented setup
Scaleway provider profileModel API for OpenClaw · Documented setup
Together AI provider profileWith models used through an API, the DigitalOcean guide runs the gateway on 1 vCPU, 1 GB of RAM and 25 GB of SSD. Local LLM inference does not fit in 1 GB.
Building the image from source needs at least 6 GB of RAM; smaller servers should use the official pre-built image.
Secure default: keep the gateway on loopback and reach it through an SSH tunnel or Tailscale Serve.
Open-source agent from Nous Research
Always-on agent gateway, formerly Clawdbot
You are on this page
Local models behind an HTTP API
High-throughput inference server
Open-source LLM app and workflow builder
Low-code builder for LLM flows and agents
Filters match a provider when one of its listed offers meets every GPU, VRAM, region, price and billing-increment condition together. Spot, reserved and serverless billing list each provider that offers that pricing and show only prices on that basis, or “not verified” when we have none. Without a GPU filter, providers are matched on their own regions and lead with their monthly plans. Read the full methodology and our affiliate disclosure.
Yes. OpenClaw's own install guides cover several VPS providers. With models used through an API, its DigitalOcean guide runs the gateway on 1 vCPU and 1 GB of RAM (accessed 26 Sep 2026).
OpenClaw's DigitalOcean guide uses 1 GB of RAM when models run through an API, and its Hetzner guide asks for at least 6 GB to build the image from source (accessed 26 Sep 2026). Running a local model on the same server needs far more memory, or a GPU. The monthly prices on this page therefore skip plans with less than 1 GB of RAM.
Yes. OpenClaw was previously called Clawdbot and then Moltbot, so older guides may use those names.
The official docs recommend keeping it on loopback and connecting through an SSH tunnel or Tailscale Serve rather than exposing it to the internet.
Some hosts sell the first term at a lower introductory price and charge a higher price when the plan renews. We rank and label plans by the renewal or standard monthly price and show the introductory price next to it, so a first-term discount is never presented as the running cost.