Hosting

25 ways to host Hermes Agent — VPS, self-hosted hardware, free tiers, and serverless. Reviews, specs, prices, and setup guides for every budget.

25
Total Options
10
VPS
9
Self-Hosted
1
Free Options
25 items
☁️ VPS Providers

Hetzner Cloud CX

VPS
★ Recommended
★★★★★
€5.49/month (CX23)

Best price-to-performance ratio in the VPS market. x86 and ARM options. Hermes runs perfectly on their cheapest CX23 instance. German hosting, excellent network.

Specs: CX23: 2 vCPU, 4GB RAM, 40GB NVMe. CX33: 4 vCPU, 8GB RAM, 80GB NVMe for €8.49
🎯 Best for: Best overall value. CX23 is enough for Hermes + API models. CX33 if you want Ollama with 3B-7B models.
Pros
  • Best price/performance in market
  • Datacenters in Germany, Finland, USA and Singapore
  • Excellent network
  • Simple control panel
  • Cheap ARM (CAX) instances too
Cons
  • No managed support on basic plans
  • Datacenters limited to EU, US and Singapore
  • Some IPs flagged by APIs initially
Setup guideSign up at hetzner.com → create project → add SSH key → create CX23 VM (Ubuntu 24.04) → ssh in → curl install hermes → setup provider → run gateway
After SSH: `curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash` then `hermes setup`
☁️ VPS Providers

Hetzner Cloud CAX (ARM)

VPS
★ Recommended
★★★★★
€5.49/month (from, CAX11)

ARM-based instances at Hetzner's lowest prices. CAX11 (2 vCPU, 4GB RAM) is perfect for running Hermes 24/7 on a budget.

Specs: CAX11: 2 ARM vCPU, 4GB RAM, 40GB NVMe. CAX21: 4 ARM vCPU, 8GB RAM, 80GB NVMe.
🎯 Best for: Cheapest reliable 24/7 Hermes host. CAX11 for API-only use, CAX21 for local 3B models.
Pros
  • Cheapest reliable VPS tier
  • Same great network as x86
  • ARM efficiency = lower power
  • Available in Germany, Finland, USA and Singapore
Cons
  • ARM-only software compatibility checks needed
  • No bare metal ARM option
  • Less CPU perf than x86 for model inference
Setup guideSame as CX series — select ARM image when creating VM. Ubuntu 24.04 ARM works perfectly.
Same as CX series. Hermes runs natively on ARM. For Ollama: `ollama pull llama3.2:3b` works on ARM.
☁️ VPS Providers

Hostinger KVM VPS

VPS
★★★★☆
$6.49/month (KVM 1, 2-year term; renews $11.99)

Mass-market host with aggressive intro pricing and a self-serve affiliate program. KVM 1 (4GB RAM, NVMe) suits API-model Hermes 24/7; KVM 2 (8GB) covers small local models.

Specs: KVM 1: 1 vCPU, 4GB RAM, 50GB NVMe, 4TB bandwidth. KVM 2: 2 vCPU, 8GB RAM, 100GB NVMe ($8.99 intro, renews $14.99). KVM 4: 4 vCPU, 16GB RAM, 200GB NVMe ($12.99 intro).
🎯 Best for: Users who want a big-brand host with AI-assisted tooling and the lowest intro price for 4GB RAM. Read the renewal price before committing.
Pros
  • Cheapest intro price for 4GB RAM
  • NVMe storage on all plans
  • Global datacenters (US, EU, APAC)
  • Weekly backups included
  • 24/7 live chat support
Cons
  • Intro price requires a 2-year commitment
  • Renewal is ~85% higher (KVM 1: $11.99/mo)
  • 1 vCPU on KVM 1 limits local-model use
  • No hourly billing — monthly minimum
Setup guideSign up at hostinger.com → choose KVM plan (intro price requires 24-month term) → pick datacenter → Ubuntu 24.04 → use Browser terminal or SSH → install Hermes
After SSH: `curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash` then `hermes setup`
☁️ VPS Providers

Contabo Cloud VPS

VPS
★★★★☆
€5.50/month (Cloud VPS 4)

German provider with the most RAM and vCPU per euro in the market. Cloud VPS 4 gives 4 vCPU and 8GB RAM for less than a DigitalOcean droplet with a quarter of that memory.

Specs: Cloud VPS 4: 4 vCPU, 8GB RAM, 100GB SSD, unlimited traffic. Page advertises "from €4.40/month".
🎯 Best for: Budget self-hosters who want 8GB RAM at a VPS price. A solid Hermes + Ollama 3B host.
Pros
  • Cheapest RAM per euro
  • 8GB RAM at a 2GB-VPS price
  • Unlimited traffic
  • EU and US datacenters
Cons
  • CPU is oversold — burst performance varies
  • Slower provisioning and support than Hetzner
  • Setup fees on some billing periods
Setup guideSign up at contabo.com → choose Cloud VPS 4 → Ubuntu 24.04 → add SSH key → deploy → SSH in → install Hermes + Ollama
Standard install: `curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash` then `hermes setup`. 8GB RAM is enough for Ollama with 3B-7B models.
☁️ VPS Providers

OVHcloud VPS

VPS
★★★★☆
$4.54/month (VPS-1)

European cloud provider with cheap NVMe-backed VPS tiers and a large EU network footprint.

Specs: VPS-1: 2 vCore, 4GB RAM, 40GB NVMe from $4.54/month. VPS-3: 6 vCore, 12GB RAM, 100GB NVMe from $12.32/month.
🎯 Best for: Cheapest 4GB VPS in this list. Good Hermes + API host; step up to VPS-3 for local models.
Pros
  • 4GB RAM for under $5/month
  • NVMe storage on entry tiers
  • Large EU datacenter footprint
Cons
  • Checkout flow less polished than DigitalOcean
  • Support tiers cost extra
  • No ARM instances on the VPS line
Setup guideSign up at ovhcloud.com → order a VPS → Ubuntu 24.04 → add SSH key → deploy → SSH in → install Hermes
Standard install: `curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash` then `hermes setup`.
☁️ VPS Providers

Scaleway Instances

VPS
★★★★☆
€4.99/month (from)

French cloud provider with hourly-billed instances starting at €4.99/month. Flexible sizing and a good EU footprint.

Specs: Instances starting at €4.99/month with hourly billing. Dedibox VPS also from €4.99/month.
🎯 Best for: EU-based Hermes hosting with per-second billing. Good when you want to resize or pause on demand.
Pros
  • Entry instances from €4.99/month
  • Per-second billing
  • Strong EU datacenter coverage
Cons
  • Console is less beginner-friendly than DigitalOcean
  • Several instance families to compare pricing across
  • Entry tier lacks the RAM for local models
Setup guideSign up at scaleway.com → Instances → create → Ubuntu 24.04 → add SSH key → start → SSH in → install Hermes
Standard install: `curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash` then `hermes setup`.
☁️ VPS Providers

DigitalOcean Basic Droplet

VPS
★★★★☆
$6/month (Basic 1GiB)

Developer-friendly VPS with excellent documentation and UI. Slightly pricier than Hetzner but easier to use. Good for beginners.

Specs: Basic: 1 vCPU, 1GB RAM, 25GB SSD, 1TB transfer for $6. 2GB: 1 vCPU, 50GB SSD, 2TB transfer for $12. 4GB: 2 vCPU, 80GB SSD, 4TB transfer for $24
🎯 Best for: Beginners who want an easy control panel. Good docs and community. Take the $12 (2GB) plan — the $6 droplet has only 1GB RAM.
Pros
  • Excellent documentation
  • Clean, intuitive dashboard
  • Good community tutorials
  • Team management features
  • Wide region selection
Cons
  • More expensive than Hetzner
  • 1GB RAM at $6 vs 4GB at €5.49 on Hetzner
  • Transfer overage costs
  • No ARM option
Setup guideDigitalOcean → Create Droplet → Ubuntu 24.04 → add SSH key → create → ssh in → install Hermes → configure gateway
Standard install. `curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash` works on all plans.
We earn commissions when you shop through the links below.View plans at DigitalOcean →
☁️ VPS Providers

Linode / Akamai Nanode

VPS
★★★★☆
$5/month (Nanode 1GB)

Akamai's cloud (formerly Linode). Solid mid-range option with good network and simple pricing. Their Nanode 1GB is cheap but minimal.

Specs: Nanode: 1 vCPU, 1GB RAM, 25GB SSD, 1TB transfer for $5. 2GB: 1 vCPU, 50GB SSD for $12. Shared 4GB: 2 vCPU, 4GB RAM, 80GB SSD, 4TB transfer for $24
🎯 Best for: Step up to the 2GB ($12) or 4GB ($24) plan for Hermes — the 1GB Nanode is too tight for anything beyond basic use.
Pros
  • Simple pricing
  • Good network quality
  • Long-established provider
  • Excellent support
  • API-first design
Cons
  • 1GB Nanode is underpowered
  • Pricier than Hetzner for equivalent spec
  • Shared 4GB plan is $24/month
  • Backups cost extra
Setup guideLinode → Create Linode → Ubuntu 24.04 → choose plan → add SSH key → deploy → SSH in → install Hermes
Standard install. The 4GB plan gives room for Ollama with 3B models. 1GB plan is API-only.
☁️ VPS Providers

Vultr Cloud Compute

VPS
★★★★☆
$10/month (2GB plan)

Global provider with 32+ datacenter locations. Good for low-latency connections worldwide. Regular plans are competitively priced.

Specs: 2GB: 1 vCPU, 2GB RAM, 55GB SSD, 2TB transfer for $10. 4GB: 2 vCPU, 4GB RAM, 80GB SSD, 3TB transfer for $20
🎯 Best for: Best for geographic diversity — deploy Hermes close to your API provider or users. The 4GB ($20) plan for local models.
Pros
  • 32+ datacenter locations (best global coverage)
  • Competitive pricing
  • ISO library
  • S3-compatible object storage
  • One-click apps
Cons
  • Slightly slower network than Hetzner
  • Support response times vary
  • Bare metal is expensive
Setup guideVultr → Deploy Server → choose location closest to you → Ubuntu 24.04 → 2GB plan ($10) → SSH key → deploy → SSH in → install Hermes
Standard install. Use the location closest to your LLM provider API endpoint for lowest latency.
We earn commissions when you shop through the links below.View plans at Vultr →
☁️ VPS Providers

AWS Lightsail

VPS
★★★☆☆
$5/month (512MB RAM)

Amazon's simplified VPS. Easier than EC2 but more expensive. Good if you're already in the AWS ecosystem. Predictable pricing.

Specs: $5: 512MB RAM, 2 vCPU, 20GB SSD, 1TB transfer. $10: 2GB RAM, 2 vCPU, 60GB SSD, 3TB transfer. $20: 4GB RAM, 2 vCPU, 80GB SSD, 4TB transfer
🎯 Best for: Only if you're already in the AWS ecosystem. Skip the $5 plan — the $10 (2GB) plan is the practical floor for Hermes.
Pros
  • Part of AWS ecosystem
  • Simple pricing (no surprise bills)
  • Good documentation
  • Built-in monitoring
  • Managed backups available
Cons
  • More expensive than Hetzner/DigitalOcean
  • 512MB $5 plan is too small for Hermes
  • Smaller SSD and less RAM per dollar than competitors
  • Every Linux plan is 2 vCPU
Setup guideAWS Console → Lightsail → Create instance → Linux/Unix → Ubuntu 24.04 → $10 plan → create → SSH via browser → install Hermes
Standard install. The browser-based SSH client in Lightsail is convenient. Ensure your instance has enough RAM (2GB+).
🏠 Self-Hosted

Raspberry Pi 5

SBC
★★★★☆
$80(board only, 8GB model)

Low-power, always-on single-board computer. Pi 5 (8GB) runs Hermes + Ollama 3B models well. Silent, cheap to run, great for learning.

Specs: ARM Cortex-A76 (4 core), 8GB LPDDR4X RAM, microSD or NVMe via HAT, USB 3.0, Gigabit Ethernet
🎯 Best for: Low-power always-on Hermes. Great for gateway + API-only use. Can run 3B models via Ollama on 8GB model.
Pros
  • Very low power (~5-15W)
  • Silent (fanless with heatsink)
  • $60-80 one-time cost
  • Great learning platform
  • GPIO for home automation integration
Cons
  • ARM64 — check software compatibility
  • Limited RAM for local models
  • microSD can be slow (use NVMe)
  • No GPU for local inference
  • Network via USB Ethernet on older models
Setup guideGet Pi 5 + 32GB+ microSD (or NVMe HAT) → flash Ubuntu Server 24.04 LTS → enable SSH → `curl ... | bash` install Hermes → configure gateway → let it run 24/7
`curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash`. For Ollama: `ollama pull llama3.2:3b`. Install the gateway as systemd service: `hermes gateway install && hermes gateway start`.
🏠 Self-Hosted

Intel NUC / Mini PC

Mini PC
★ Recommended
★★★★★
$150-300(used/refurbished)

Small form-factor x86 PC. Much more powerful than Pi for local models. Can run 7B-13B models with decent speed. Silent or near-silent.

Specs: Intel i3/i5/i7 (8th-13th gen), 8-32GB RAM, 256GB+ NVMe SSD. Fanless cases available for some models.
🎯 Best for: Best self-hosted option. NUC with 16GB RAM can run 7B models comfortably. Refurbished units are excellent value.
Pros
  • Full x86 compatibility
  • Can run 7B-13B models
  • Silent or near-silent
  • Low power (~15-30W)
  • Small footprint
  • Refurbished units are cheap
Cons
  • Higher upfront cost than Pi
  • Need to source used/refurbished
  • Some models have fan noise
  • Not as portable as Pi
Setup guideGet a used NUC (eBay: $150 for 8th gen i5) → add NVMe SSD + RAM → install Ubuntu Server 24.04 → SSH in → install Hermes → install Ollama → configure gateway + cron jobs
Full install with Ollama for local models: `curl -fsSL https://ollama.com/install.sh | sh` → `ollama pull llama3.2` (for API) or larger models. Hermes config: add a Custom Endpoint with base_url http://localhost:11434/v1 (no key). Run gateway as service.
🏠 Self-Hosted

Old Laptop / Repurpose

Repurpose
★ Recommended
★★★★★
$0(free, you already own it)

Best bang for buck — that old laptop in a drawer is a perfect Hermes server. Built-in UPS (battery), WiFi, keyboard for emergency access.

Specs: Any laptop with 8GB+ RAM, SSD preferred. 4GB is usable for API-only. Intel 8th gen or newer recommended.
🎯 Best for: Ultimate budget option — it's free. Even a 5-year-old laptop is overkill for Hermes + API models.
Pros
  • Completely free (you own it)
  • Built-in UPS (battery backup)
  • WiFi + Ethernet
  • Can also serve as media server
  • Good performance (i5/i7)
Cons
  • Takes up desk space
  • Screen is wasted (headless)
  • Fan noise
  • Older models use more power
  • Not as "clean" as a dedicated solution
Setup guideInstall Ubuntu Server (or keep existing OS) → plug in Ethernet for reliability → close lid (configure to stay on when closed) → install Hermes → install Ollama if desired → run gateway → forget it exists
Same as NUC. Close-lid behavior: edit `/etc/systemd/logind.conf` → set `HandleLidSwitch=ignore` → `systemctl restart systemd-logind`. Run Hermes gateway headless.
🏠 Self-Hosted

Home Server / Tower

Server
★★★★☆
$300-800(build or buy)

Dedicated server hardware at home. Overkill for Hermes alone but great if you're running other services (NAS, Plex, Home Assistant).

Specs: Any modern CPU, 16-64GB RAM, SSD storage. Can run 13B-70B models locally. Supports GPU inference with NVIDIA card.
🎯 Best for: Power users who already run a home server. Hermes is a lightweight addition to an existing Proxmox/Docker setup.
Pros
  • Maximum performance
  • GPU support for local models
  • Run with other services
  • Full control over hardware
  • Can handle 70B models with decent GPU
Cons
  • Expensive
  • High power consumption
  • Loud (fans)
  • Takes up space
  • Overkill for Hermes alone
Setup guideInstall Proxmox or Ubuntu Server → create VM or Docker container for Hermes → install via script → configure gateway → add to startup → manage alongside other services
Docker: `docker run -d --name hermes -v ~/.hermes:/opt/data nousresearch/hermes-agent`. Or native install. GPU passthrough for local LLM inference.
🏠 Self-Hosted

NAS (Synology / QNAP)

NAS
★★★☆☆
$0(if you already have a NAS)

Run Hermes on your existing NAS via Docker. Most modern NAS units have enough power. Uses storage you already have.

Specs: Intel Celeron or better, 4GB+ RAM, Docker support. Note the Synology DS220+ ships with 2GB RAM (expandable) and Synology has dropped Container Manager on some recent consumer models — verify both per model. QNAP TS-2xx or better.
🎯 Best for: If you already run a NAS, adding Hermes is free. Good for gateway + cron jobs. Not for local models.
Pros
  • Uses existing hardware
  • 24/7 operation (NAS runs anyway)
  • RAID storage redundancy
  • Low additional power use
Cons
  • Limited RAM (4-8GB typical; DS220+ ships with 2GB)
  • Underpowered CPU
  • No GPU
  • Docker on NAS can be finicky
  • Not for local models
Setup guideNAS → Package Center → install Docker/Container Manager → pull hermes-agent image → create container with volume mounts for config → set up port/gateway → start
Docker-based: `docker run -d --name hermes -v /volume1/docker/hermes:/opt/data nousresearch/hermes-agent:latest`. Configure via the mounted config volume.
🏠 Self-Hosted

Hardware: Absolute Cheapest 24/7

SBC
★★★☆☆
$15-30(one-time, used hardware)

Run Hermes on a used thin client from eBay, or an old Android phone via Termux (best-effort, Tier-2). The absolute lowest-cost always-on setup.

Specs: Any old Android phone (4GB RAM+) or thin client (Wyse, HP t620). ARM or x86. USB power for the phone.
🎯 Best for: Absolute budget builds. $20 thin client from eBay + $5 USB power adapter = cheapest 24/7 Hermes server. For Android, use the documented Termux installer, not a bare pip install.
Pros
  • Extremely cheap
  • Low power (phone: 2-5W)
  • Teaches Linux skills
  • Repurposes e-waste
Cons
  • Limited performance
  • Setup is more involved
  • Phone battery swelling risk if always plugged
  • Thin clients have weak CPUs
  • Termux is Tier-2 best-effort: no Docker isolation and Android may suspend background jobs
Setup guideThin client: install Linux (Debian/Ubuntu) on internal storage or USB → SSH → install Hermes. Android: follow the Termux installer + venv guide (bit.ly path in docs), then run the gateway.
On thin client: standard Linux install. On Android/Termux: run the Termux-aware installer script + venv from the docs — Docker-based terminal isolation is unavailable and gateway persistence is best-effort.
🏠 Self-Hosted

Singularity / Apptainer

Server
★★★☆☆
$0(software; you supply the HPC allocation)

One of the seven Hermes terminal backends, aimed at HPC and shared machines where Docker or root access is unavailable. Containers run unprivileged from a single image file.

Specs: Runs on HPC clusters and shared multi-user machines without root. Selected with terminal.backend singularity.
🎯 Best for: Shared cluster and HPC environments where you cannot run a Docker daemon.
Pros
  • Works without root or a Docker daemon
  • Common on HPC systems
  • Single-file container images
Cons
  • Not available on managed VPS hosts
  • Unprivileged model limits some features
  • Setup varies per cluster
Setup guideInstall Singularity or Apptainer → pull or build a container image → `hermes config set terminal.backend singularity` → run Hermes
Set `terminal.backend` to `singularity` in config.yaml, then ensure the container image is available on the host.
🏠 Self-Hosted

Hermes Desktop

Mini PC
★★★★☆
$0(free installer)

The recommended macOS and Windows installer path. `hermes desktop` installs and launches the Electron desktop app, which bundles the agent core with a GUI, embedded terminal panes, Projects, and the desktop-UI tools.

Specs: Electron desktop app for macOS and Windows. Includes GUI session panes, Projects, and the desktop-only toolsets.
🎯 Best for: Desktop users who want a guided install and a GUI instead of running the CLI by hand.
Pros
  • One-command install on macOS and Windows
  • GUI plus embedded terminal panes
  • Full desktop toolset available
Cons
  • Desktop machine must stay awake for gateway duty
  • Not the path for headless servers
  • GUI-only tools do not exist in CLI sessions
Setup guideRun `hermes desktop` → the installer sets up the app and agent core → sign in and start a session from the GUI
`hermes desktop` installs and launches the desktop app. The same data directory is used, so CLI and desktop sessions share config, memory, and profiles.
🏠 Self-Hosted

Official Docker Image

Docker
★★★★☆
$0(image is free; you pay for the host)

The officially published Hermes container image, `nousresearch/hermes-agent:latest`, with Docker Compose files in the repository. The docs devote a full page to container deployments.

Specs: Prebuilt image with the agent core and dependencies. Run standalone or via the bundled docker-compose files, mounting ~/.hermes as a volume. The entrypoint is a dispatcher that handles PID-1 wrapping for platforms like Fly.io.
🎯 Best for: Reproducible, isolated deployments on any Docker host - the standard path for self-hosting on a NAS, home server, or VPS.
Pros
  • Reproducible across hosts
  • Easy upgrades by pulling a new tag
  • Compose files provided for multi-service setups
Cons
  • Needs a Docker host
  • Volume mounts are required to persist state
  • Container networking adds gateway setup complexity
Setup guideInstall Docker → `docker pull nousresearch/hermes-agent:latest` → run with `~/.hermes` mounted at `/opt/data` → configure the gateway inside the container
`docker run -d --name hermes -v ~/.hermes:/opt/data nousresearch/hermes-agent:latest`, or use the repository docker-compose.yml.
🆓 Free Tier

Oracle Cloud Free Tier

Cloud
★★★★☆
$0/month (always free)

Always-free instances with no time limit: 2 ARM OCPUs / 12GB RAM (Ampere A1) plus 2 AMD x86 micro instances. The best free hosting for a persistent Hermes gateway.

Specs: Always Free: 2 ARM OCPUs (Ampere A1) + 12GB RAM, 200GB block volume total, 2 x AMD VM.Standard.E2.1.Micro (x86), 20GB object storage.
🎯 Best for: Best free option for an always-on gateway. 12GB RAM is plenty for Hermes with API models and small local models via Ollama.
Pros
  • 12GB RAM free forever
  • 2 ARM OCPUs plus 2 AMD micro instances
  • 200GB block volume
  • No time limit (always free)
Cons
  • Oracle signup can be finicky
  • Free ARM capacity is often unavailable in busy regions
  • AMD micro instances are x86 but very small
  • Support is paid-only
  • Oracle dashboard is complex
Setup guideSign up at oracle.com/cloud/free (may need a card for verification) → create VM → select Ampere A1 (2 OCPU / 12GB) or a free AMD micro shape → Ubuntu 24.04 → SSH in → install Hermes
Standard install. Then install Ollama: `curl -fsSL https://ollama.com/install.sh | sh` → `ollama pull llama3.2:3b` on the 12GB ARM instance. Point Hermes at it via the Custom Endpoint flow (base_url http://localhost:11434/v1, no key).
⚡ Serverless

Google Cloud Shell

Cloud
★★☆☆☆
$0(free, 50h/week quota)

Free browser-based terminal with a 5GB persistent home directory. Great for trying Hermes instantly with zero setup. The VM is ephemeral, but files in $HOME survive between sessions.

Specs: Free: 5GB persistent $HOME, ephemeral VM (discarded after 1h idle, 12h hard cap). Pre-installed with Python, Node and git.
🎯 Best for: Quick testing and evaluation: the home directory persists, so an install survives sessions. Not for production — the VM is discarded and ports are not exposed.
Pros
  • Completely free
  • Persistent 5GB home directory
  • No installation on your machine
  • Pre-installed tools
  • Accessible from any browser
Cons
  • VM is ephemeral (1h idle timeout, 12h hard cap)
  • 50 hours/week quota
  • No gateway (ports not exposed)
  • Intended for interactive use only
  • Google account needed
  • $HOME deleted after 120 days of inactivity
Setup guideGo to shell.cloud.google.com → `curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash` → `hermes setup` → try it. Sessions end after 1h idle (12h hard cap); $HOME persists.
Direct curl install. No gateway available. Best for `hermes chat -q` one-off queries during testing.
⚡ Serverless

Modal

Cloud
★★★☆☆
Free tier($30/month free credits)

Serverless GPU/CPU compute. Hermes has native Modal support as a terminal backend. Good for burst inference with GPU. Free tier available.

Specs: Free tier: $30/month credits. CPU and GPU instances. Pay-as-you-go beyond free credits. No always-on — cold starts.
🎯 Best for: Burst inference with GPU access. Not for persistent gateway. Good for running large models (70B) on demand.
Pros
  • GPU access on free tier
  • No server management
  • Pay only for compute used
  • Native Hermes support
Cons
  • Cold starts (~10-20s)
  • No persistent gateway
  • Credit-based, not for 24/7 use
  • Learning curve for Modal SDK
Setup guideSign up at modal.com → install modal CLI → authenticate → configure Hermes with terminal backend: `modal` → `hermes chat -q "..."` runs on Modal infra
Set terminal backend: `hermes config set terminal.backend modal`. The `hermes setup` wizard has a Modal option.
⚡ Serverless

Daytona

Cloud
★★★☆☆
$200 free credit(then usage-based)

Development environment manager usable as a Hermes terminal backend. Runs isolated cloud workspaces for agent shell commands.

Specs: Usage-based: $0.0504/vCPU-hr, $0.0162/GiB-hr memory, $0.000108/GiB-hr storage, with $200 in free compute included. Hermes integration caps workspace disks at 10 GiB.
🎯 Best for: Development workflows where you need an isolated shell alongside your code. Less suited for persistent gateway hosting.
Pros
  • $200 in free compute to start
  • Isolated cloud workspaces
  • Git-integrated development environments
  • Documented Hermes terminal backend
Cons
  • Usage-based billing after the free credit
  • Not for 24/7 gateway
  • Workspace-based (not always-on)
  • 10 GiB disk cap in the Hermes integration
Setup guidedaytona.io → sign up → create workspace → set `terminal.backend: daytona` with `DAYTONA_API_KEY` → run Hermes normally
Add `DAYTONA_API_KEY`, then `hermes config set terminal.backend daytona`. Shell commands execute in a Daytona workspace.
⚡ Serverless

Vercel Sandbox

Cloud
★★★☆☆
Varies(usage-based)

First-class Hermes terminal backend that runs the agent shell in an ephemeral Vercel-managed microVM. Supports node24, node22, and python3.13 runtimes.

Specs: Ephemeral microVM sandboxes with node24 / node22 / python3.13 runtimes. Selected with terminal.backend vercel_sandbox.
🎯 Best for: Isolated execution of agent shell commands without managing a server. Good when you want per-session sandboxing rather than a persistent host.
Pros
  • Ephemeral per-session isolation
  • Multiple language runtimes
  • No server to manage
Cons
  • Usage-based billing
  • Not suitable for a persistent gateway
  • Requires Vercel credentials
Setup guideConfigure a Vercel account and credentials → `hermes config set terminal.backend vercel_sandbox` → run Hermes normally; shell commands execute in the sandbox
Set `terminal.backend` to `vercel_sandbox` in config.yaml. The `hermes setup` wizard can select the backend too.
⚡ Serverless

Fly.io Machines

Cloud
★★★☆☆
Varies(usage-based)

Container host referenced by the Hermes docs as a supported deployment target - Fly Machines run the official Docker image with a proper PID-1 entrypoint.

Specs: Fly Machines run the Hermes container image. Needs a correct PID-1 / init entrypoint in the container command.
🎯 Best for: Hosting the gateway as a container close to users, with scale-to-zero when idle.
Pros
  • Global anycast network
  • Scale to zero when idle
  • Real volumes for persistent state
Cons
  • Requires the correct container entrypoint / PID-1 handling
  • Usage-based billing
  • More moving parts than a plain VPS
Setup guideInstall flyctl → create a Fly app → deploy the Hermes image with a persistent volume for ~/.hermes → set secrets for API keys → start the gateway
Deploy the documented Hermes container image from Fly, setting API keys as Fly secrets and mounting a volume for ~/.hermes.
🔍

No results found. Try a different search or filter.