Fly.io
Always-on or scale-to-zero AI agents (OpenClaw, Hermes Agent) that call external LLM APIs
CPU instance · Fly Machine shared-cpu-1x, 256 MB · from:
$2.19≈ €1.96 (estimated conversion from $2.19)
Compare AI agent hosting: VPS, managed platforms and GPU servers
We may earn commissions when you shop through the links below. Learn more
Compare managed and self-hosted options for persistent agents, tool use, browser workloads, scheduled jobs and API-heavy automation.
AI agent hosting providers
What to check before you pick a host
- Where the model runs
Agents that call a hosted model API need CPU, memory and storage. A local model on the same machine usually needs a GPU with enough VRAM for its weights and context; small quantised models can run on the CPU, much more slowly.
- Persistence
Agents keep state, memory and workspace files. Use a persistent volume and back it up; container disks are often wiped on restart.
- Access and secrets
Keep admin interfaces off the public internet, for example behind an SSH tunnel or a private network, and store API keys as secrets.
- Always-on cost
A 24/7 agent runs about 730 hours a month. Compare monthly totals, and the price after any introductory term, not only hourly rates.
Compare by framework
- Compare Hermes Agent hosting
Hermes Agent
Open-source agent from Nous Research
- Compare OpenClaw hosting
OpenClaw
Always-on agent gateway, formerly Clawdbot
- Compare Ollama hosting
Ollama
Local models behind an HTTP API
- Compare vLLM hosting
vLLM
High-throughput inference server
- Compare managed hosting for Dify
Dify
Open-source LLM app and workflow builder
- Compare managed hosting for Flowise
Flowise
Low-code builder for LLM flows and agents
Tools for this decision
- Agent Hosting SelectorAnswer a few questions about your agent and get a best-fit hosting category and provider shortlist.
- GPU Cost CalculatorEstimate monthly GPU spend from verified hourly prices, usage hours, instances and storage.
- VRAM / Model FitEstimate the GPU memory a model needs for inference or fine-tuning, and which GPUs fit.
Explore related hubs
How we compare
Monthly plan prices.
Rows lead with the provider's cheapest verified monthly plan of this kind; with a region or EU filter, the cheapest plan available there. An introductory price is only shown next to its renewal price and is never used to rank. Prices exclude VAT and sales tax.Not verified means not verified.
When no official source confirms a value we show “Not verified” instead of an estimate. Where a provider's terms do not allow its prices to be republished, we link to its live pricing instead.Freshness on every row.
Updated today: checked on the current day (UTC). Fresh: up to 7 days. Stale: up to 30 days. Needs refresh: older than that.No paid placement.
Results only change order with the sort you pick. Affiliate relationships never affect order, filters or which providers appear.
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.
Frequently asked questions
Can I host an AI agent on a VPS?
Yes. Agents that call a hosted model API mostly need CPU, memory and persistent storage, which a VPS provides. OpenClaw's documentation, for example, runs its gateway on 1 vCPU and 1 GB of RAM when models are used through an API (accessed 26 Sep 2026).
How do I run my own AI agent 24/7?
Run it as a long-lived process or container on a server that stays online, keep its state on persistent storage with backups, and keep its admin interface private. Our guide to hosting an AI agent 24/7 covers sizing, hardening and backups step by step.
Do I need a GPU to run an AI agent?
Only if the model runs on the same machine, and even then not always. A hosted model API needs no GPU; a local model usually needs a GPU with enough VRAM for its weights plus the context window your agent uses. Small quantised models can run on a CPU, much more slowly.
What is managed agent hosting?
The provider runs the agent runtime for you, including updates, restarts, backups and scaling. In exchange you get less control over the server and depend on the platform's limits.
Why do some plans show an introductory and a renewal price?
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.
Why do some values say “Not verified”?
We only show a value when an official provider source confirms it. If we could not confirm a price or feature, the row says “Not verified” instead of showing an estimate, and the provider page lists the sources we did check.