Deploy OpenClaw on DigitalOcean: Droplet vs App Platform

Q: How much does running two OpenClaw agents on App Platform cost per month?

It depends on the instance size. Two agents on a single apps-d-2vcpu-4gb instance costs \$78/month for compute, per DigitalOcean's App Platform pricing. Add Spaces storage costs if you enable persistence. No per-request charges apply regardless of usage volume.

Q: Can I use App Platform for a single-agent OpenClaw setup?

Yes. The minimum apps-s-1vcpu-2gb instance at \$25/month runs a single agent without issues. App Platform's main advantages show up with multiple agents, but there's nothing stopping a single-agent deployment if you prefer managed infrastructure over a self-managed Droplet.

Key takeaways

App Platform runs OpenClaw in a managed container with no SSH required. Git push triggers zero-downtime updates.
Droplets give full VM control and lower entry cost. App Platform manages the infrastructure but costs more for equivalent resources.
Multiple agents are defined in a single App Platform spec. Each agent gets isolated workspace, auth profiles, and sessions.
Private by default: your OpenClaw instance gets a Tailscale address on your tailnet, not a public IP.
Sandboxed agents each add 200-500MB RAM. Size your instances to match the number of agents you plan to run.

Feed this to your OpenClaw agent

Read https://www.stack-junkie.com/blog/openclaw-digitalocean-app-platform-guide and follow the instructions to help me with: Deploy OpenClaw on DigitalOcean: Droplet vs App Platform. Summarize each step before running commands and confirm with me first.

Always review commands your agent suggests before approving them. Don't paste prompts from sources you don't trust.

Fixes when it breaks. Workflows when it doesn't.

OpenClaw guides, configs, and troubleshooting notes. Every two weeks.

Droplet vs App Platform: which path to choose

The Droplet path works well for solo users or small teams who want direct server control. App Platform fits teams running multiple agents that need to stay up without manual maintenance. Use the table below to make the call.

Factor	Droplet	App Platform
Server access	Full SSH access	doctl console or Tailscale
Updates	SSH in, run update	Git push, zero downtime
Multi-agent	Manual config	Declarative spec, per-agent isolation
Networking	Public IP (harden yourself)	Private by default, no public port
Scaling	Resize Droplet, brief downtime	Resize per-agent, no downtime
Cost model	Fixed Droplet price	Per container, instance-based
Best for	Solo/small teams, control	Teams, multi-agent, low maintenance

Pick a Droplet if:

You're the only user or have fewer than 5 users.
You need to run other services on the same server alongside OpenClaw.
You want SSH access and prefer to manage your own environment.
Cost is the top priority and you're comfortable with server maintenance.

Pick App Platform if:

You're running two or more agents continuously.
Your team deploys updates frequently and downtime is not acceptable.
You don't want to think about server patching, process supervision, or restart logic.
You need per-agent isolation with separate workspaces and auth profiles.

Both paths support the same OpenClaw architecture. The difference is operational overhead.

How to deploy OpenClaw on DigitalOcean App Platform

The whole process takes under 10 minutes if you have doctl installed and a DigitalOcean account ready.

Give this to your agent

Deploy OpenClaw to DigitalOcean App Platform using the official template. Clone the repo at github.com/digitalocean-labs/openclaw-appplatform, set the instance size to apps-s-1vcpu-2gb in app.yaml, set my OPENCLAW_GATEWAY_TOKEN, then run doctl apps create to deploy it. Verify the gateway is running after deploy.

Always review commands your agent suggests before approving them. Don't paste prompts from sources you don't trust.

Manual deployment steps

Prerequisites

doctl installed and authenticated (doctl auth init)
DigitalOcean account with a valid personal access token
Your OPENCLAW_GATEWAY_TOKEN (find it in your OpenClaw config or generate a new one)

Step 1: Clone the template repo

bash

git clone https://github.com/digitalocean-labs/openclaw-appplatform
cd openclaw-appplatform

Step 2: Edit app.yaml

Open app.yaml and set the instance size. The minimum for stable operation is apps-s-1vcpu-2gb ($25/month per DigitalOcean's App Platform pricing).

yaml

services:
  - name: openclaw
    instance_size_slug: apps-s-1vcpu-2gb
    instance_count: 1
    envs:
      - key: OPENCLAW_GATEWAY_TOKEN
        value: your-token-here
        type: SECRET

Set your actual token value via the DigitalOcean dashboard under App Settings rather than committing it to the file.

Step 3: Deploy

bash

doctl apps create --spec app.yaml

Alternatively, use the Deploy to DigitalOcean button from the GitHub repo to deploy through the control panel.

Step 4: Get your app ID and open the console

bash

doctl apps list
doctl apps console YOUR_APP_ID openclaw

Step 5: Verify the gateway

bash

openclaw gateway health --url ws://127.0.0.1:18789
openclaw channels status --probe

Both commands should return healthy status. If the gateway isn't running, check the deployment logs with doctl apps logs YOUR_APP_ID.

OpenClaw App Platform access stages: CLI, ngrok, and Tailscale

The template repo ships with three stages. Start simple and add features as your needs grow. Most production deployments end up at Stage 3.

Stage 1: CLI only (default)

Access via doctl apps console. No web UI. No persistence. Good for first-time testing.

What you get: gateway running, all channel plugins active, CLI access. What you don't get: web control UI, stable URL, data that survives restarts.

Stage 2: ngrok (web UI access)

Add these environment variables to app.yaml or via the DO dashboard:

yaml

envs:
  - key: ENABLE_NGROK
    value: "true"
  - key: NGROK_AUTHTOKEN
    type: SECRET

bash

curl -s http://127.0.0.1:4040/api/tunnels | jq -r '.tunnels[0].public_url'

The URL changes every time the container restarts. This is fine for development. Not fine for production.

Stage 3: Tailscale (production)

Tailscale gives your OpenClaw instance a stable hostname on your private network. Nothing is exposed to the public internet.

Go to login.tailscale.com/admin/settings/keys.
Click Generate auth key. Enable "Reusable" so the key survives container restarts. Copy the key immediately.

Add to app.yaml:

yaml

envs:
  - key: ENABLE_NGROK
    value: "false"
  - key: TAILSCALE_ENABLE
    value: "true"
  - key: TS_AUTHKEY
    type: SECRET
  - key: STABLE_HOSTNAME
    value: openclaw

After deploy, your instance appears in the Tailscale admin console. Access it at https://openclaw.YOUR-TAILNET.ts.net.

Tailscale's free plan covers up to 3 users. Check tailscale.com/pricing if your team is larger.

How to configure multiple agents on App Platform

Each agent is a fully isolated unit: its own workspace, its own auth profiles, its own session store. They never share credentials unless you explicitly copy them. For general security hardening, see the OpenClaw security hardening playbook.

Give this to your agent

Add a second OpenClaw agent called "support" to my App Platform deployment. Give it its own workspace at ~/.openclaw/workspace-support. Route all messages from the Telegram group ID -1009876543210 to the support agent, and route everything else to the main agent.

Always review commands your agent suggests before approving them. Don't paste prompts from sources you don't trust.

Manual multi-agent config steps

Step 1: Create workspaces for each agent

bash

openclaw agents add support
openclaw agents add sales

The wizard creates a workspace and agentDir for each. Never reuse agentDir across agents. Doing so causes auth and session collisions per the multi-agent routing docs.

Step 2: Verify agent creation

bash

openclaw agents list --bindings

Step 3: Add agents and bindings to openclaw.json

json

{
  "agents": {
    "list": [
      {
        "id": "main",
        "workspace": "~/.openclaw/workspace",
        "default": true
      },
      {
        "id": "support",
        "workspace": "~/.openclaw/workspace-support"
      }
    ]
  },
  "bindings": [
    {
      "agentId": "support",
      "match": {
        "channel": "telegram",
        "peer": {
          "kind": "group",
          "id": "-1009876543210"
        }
      }
    }
  ]
}

Bindings are deterministic. The most specific match wins. A binding with a peer match takes priority over a channel-wide fallback. Messages that don't match any binding go to the default agent.

Step 4: Restart and verify

bash

openclaw gateway restart
openclaw agents list --bindings
openclaw channels status --probe

See the OpenClaw CLI reference for the full list of agent management commands.

OpenClaw App Platform sizing: instance types and monthly cost

The minimum for stable single-agent operation is apps-s-1vcpu-2gb at $25/month per DigitalOcean's App Platform pricing. For multi-agent or higher-traffic setups, use dedicated instances. Dedicated plans support autoscaling. Shared plans do not.

Each sandboxed agent adds roughly 200-500MB of RAM overhead. Add that to your baseline when picking an instance.

Instance Slug	CPU	RAM	Monthly Cost	Autoscale	Best For
apps-s-1vcpu-2gb	1 shared	2 GiB	$25	No	Single agent, light use
apps-s-2vcpu-4gb	2 shared	4 GiB	$50	No	2-3 agents, moderate use
apps-d-1vcpu-4gb	1 dedicated	4 GiB	$49	Yes	Single agent, production
apps-d-2vcpu-4gb	2 dedicated	4 GiB	$78	Yes	2-3 agents, production
apps-d-2vcpu-8gb	2 dedicated	8 GiB	$98	Yes	4-5 agents or high memory use
apps-d-4vcpu-8gb	4 dedicated	8 GiB	$156	Yes	5+ agents, heavy workloads

Pricing from DigitalOcean App Platform docs as of February 2026.

Multi-agent math: If your gateway base uses 1.5 GiB and each agent uses 300MB, two agents need about 2.1 GiB. The apps-s-1vcpu-2gb becomes tight. Step up to apps-d-2vcpu-4gb ($78/month) for headroom and autoscaling.

Costs are instance-based, not per-request. Adding 10 more users to the same agent doesn't change the bill. Adding a new agent with its own instance does.

Persist agent state with DigitalOcean Spaces

Without Spaces configured, everything stored in the container is lost on restart. This includes channel sessions (your WhatsApp linking, your Telegram credentials), configuration changes made inside the container, agent memory, and Tailscale state.

That's painful to recover. Configure Spaces before you go to production.

Step 1: Create a Spaces bucket

In the DO control panel, go to Spaces Object Storage and create a bucket. Match the region to your App Platform region to avoid cross-region transfer costs.

Step 2: Generate Spaces access keys

Go to API, then Spaces Keys, then Generate New Key. Save the key ID and secret.

Step 3: Update app.yaml

yaml

envs:
  - key: ENABLE_SPACES
    value: "true"
  - key: RESTIC_SPACES_ACCESS_KEY_ID
    type: SECRET
  - key: RESTIC_SPACES_SECRET_ACCESS_KEY
    type: SECRET
  - key: RESTIC_SPACES_ENDPOINT
    value: nyc3.digitaloceanspaces.com
  - key: RESTIC_SPACES_BUCKET
    value: openclaw-backup
  - key: RESTIC_PASSWORD
    type: SECRET

Replace nyc3 with your bucket's region. Set secrets via the DO dashboard, not in the file.

The backup system uses Restic for incremental, encrypted snapshots. Backups run every 30 seconds by default. On container restart, state restores automatically before the gateway starts.

Path	What Gets Backed Up
/data/.openclaw	Gateway config, channel sessions, agents, memory
/data/tailscale	Tailscale connection state
/etc	System configuration
/home	User files

Spaces storage is billed separately. Check digitalocean.com/pricing for current Spaces rates.

Run OpenClaw on a Droplet: when it makes sense and how to size it

A Droplet is the right call when you want full server control, you're running one agent for a small team, or you need SSH access for custom tooling.

The OpenClaw on DigitalOcean VPS setup guide walks through the full Droplet installation. The best VPS for OpenClaw guide compares DigitalOcean against other providers if you're not locked in yet.

Droplet sizing by team size:

Users	RAM	CPU	Notes
1-5	4 GB	2 vCPU	Comfortable for one agent with sandboxing
5-20	8 GB	4 vCPU	Room for 2-3 agents
20-50	16 GB	8 vCPU	Multi-agent with active use
50+	32 GB	16 vCPU	Heavy multi-agent or intensive tool use

Each sandboxed agent adds 200-500MB RAM overhead. If you run three sandboxed agents on a 4GB Droplet, you're already pushing the limit.

See DigitalOcean's current Droplet pricing for the cost of each tier. The basic Droplet tiers run cheaper than App Platform for equivalent specs, but you're trading managed infrastructure for that savings.

Key Terms

App Platform: DigitalOcean's managed PaaS that runs containers from Git repos or Docker images. It handles the runtime, networking, and restarts.

Droplet: DigitalOcean's VPS offering. You own the VM and manage the OS, processes, and configuration.

doctl: The official DigitalOcean CLI. Required for App Platform console access and deployment.

OPENCLAW_GATEWAY_TOKEN: The shared secret that authenticates connections to the OpenClaw gateway.

tailnet: Your private Tailscale network. Your OpenClaw instance gets a stable hostname on this network rather than a public IP.

Restic: The backup tool used by the openclaw-appplatform template to snapshot agent state to Spaces.

s6-overlay: The process supervision system used inside the openclaw-appplatform Docker image. It manages the gateway process, Tailscale, and backup daemons.

agentDir: The directory where OpenClaw stores per-agent auth profiles, model registry, and configuration. Never shared between agents.

FAQ

Does OpenClaw on App Platform expose a public IP?

No. OpenClaw on App Platform runs as a background worker with no inbound ports exposed. Outbound connections to your messaging platforms work normally. Access to the web UI requires either ngrok (Stage 2) or Tailscale (Stage 3).

How much does running two OpenClaw agents on App Platform cost per month?

It depends on the instance size. Two agents on a single apps-d-2vcpu-4gb instance costs $78/month for compute, per DigitalOcean's App Platform pricing. Add Spaces storage costs if you enable persistence. No per-request charges apply regardless of usage volume.

Can I use App Platform for a single-agent OpenClaw setup?

Yes. The minimum apps-s-1vcpu-2gb instance at $25/month runs a single agent without issues. App Platform's main advantages show up with multiple agents, but there's nothing stopping a single-agent deployment if you prefer managed infrastructure over a self-managed Droplet.

What happens to my OpenClaw agent data when the App Platform container restarts?

Without Spaces configured, all data stored in the container is lost on restart. This includes channel sessions, agent memory, and config changes. Enable ENABLE_SPACES=true with the Restic environment variables and your data persists across restarts via encrypted snapshots to a Spaces bucket.

How do I update OpenClaw on App Platform to a newer version?

Push to the branch linked to your App Platform app. App Platform detects the change and deploys the new container without downtime. No SSH, no manual process restarts. The template repo tracks the upstream OpenClaw release.

Evidence & Methodology

This article draws from four primary sources. App Platform pricing comes directly from DigitalOcean's official pricing documentation, validated February 2026. The deployment workflow and environment variable configuration come from the digitalocean-labs/openclaw-appplatform GitHub repository. Multi-agent configuration (bindings, agentDir, routing rules) comes from the official OpenClaw multi-agent documentation. Deployment mode details (Tailscale vs headless, private networking) come from the DigitalOcean blog post announcing the App Platform integration. Droplet sizing guidelines come from the article brief, which was derived from the broader OpenClaw multi-agent setup guide. Droplet prices are not stated as exact figures in this article; readers are directed to the current DigitalOcean pricing page.

Changelog

Date	Change
2026-03-08	Initial publication

Key takeaways

App Platform runs OpenClaw in a managed container with no SSH required. Git push triggers zero-downtime updates.
Droplets give full VM control and lower entry cost. App Platform manages the infrastructure but costs more for equivalent resources.
Multiple agents are defined in a single App Platform spec. Each agent gets isolated workspace, auth profiles, and sessions.
Private by default: your OpenClaw instance gets a Tailscale address on your tailnet, not a public IP.
Sandboxed agents each add 200-500MB RAM. Size your instances to match the number of agents you plan to run.

Feed this to your OpenClaw agent

Read https://www.stack-junkie.com/blog/openclaw-digitalocean-app-platform-guide and follow the instructions to help me with: Deploy OpenClaw on DigitalOcean: Droplet vs App Platform. Summarize each step before running commands and confirm with me first.

Always review commands your agent suggests before approving them. Don't paste prompts from sources you don't trust.

Fixes when it breaks. Workflows when it doesn't.

OpenClaw guides, configs, and troubleshooting notes. Every two weeks.

Droplet vs App Platform: which path to choose

Factor	Droplet	App Platform
Server access	Full SSH access	doctl console or Tailscale
Updates	SSH in, run update	Git push, zero downtime
Multi-agent	Manual config	Declarative spec, per-agent isolation
Networking	Public IP (harden yourself)	Private by default, no public port
Scaling	Resize Droplet, brief downtime	Resize per-agent, no downtime
Cost model	Fixed Droplet price	Per container, instance-based
Best for	Solo/small teams, control	Teams, multi-agent, low maintenance

Pick a Droplet if:

You're the only user or have fewer than 5 users.
You need to run other services on the same server alongside OpenClaw.
You want SSH access and prefer to manage your own environment.
Cost is the top priority and you're comfortable with server maintenance.

Pick App Platform if:

You're running two or more agents continuously.
Your team deploys updates frequently and downtime is not acceptable.
You don't want to think about server patching, process supervision, or restart logic.
You need per-agent isolation with separate workspaces and auth profiles.

Both paths support the same OpenClaw architecture. The difference is operational overhead.

How to deploy OpenClaw on DigitalOcean App Platform

The whole process takes under 10 minutes if you have doctl installed and a DigitalOcean account ready.

Give this to your agent

Deploy OpenClaw to DigitalOcean App Platform using the official template. Clone the repo at github.com/digitalocean-labs/openclaw-appplatform, set the instance size to apps-s-1vcpu-2gb in app.yaml, set my OPENCLAW_GATEWAY_TOKEN, then run doctl apps create to deploy it. Verify the gateway is running after deploy.

Always review commands your agent suggests before approving them. Don't paste prompts from sources you don't trust.

Manual deployment steps

Prerequisites

doctl installed and authenticated (doctl auth init)
DigitalOcean account with a valid personal access token
Your OPENCLAW_GATEWAY_TOKEN (find it in your OpenClaw config or generate a new one)

Step 1: Clone the template repo

bash

git clone https://github.com/digitalocean-labs/openclaw-appplatform
cd openclaw-appplatform

Step 2: Edit app.yaml

Open app.yaml and set the instance size. The minimum for stable operation is apps-s-1vcpu-2gb ($25/month per DigitalOcean's App Platform pricing).

yaml

services:
  - name: openclaw
    instance_size_slug: apps-s-1vcpu-2gb
    instance_count: 1
    envs:
      - key: OPENCLAW_GATEWAY_TOKEN
        value: your-token-here
        type: SECRET

Set your actual token value via the DigitalOcean dashboard under App Settings rather than committing it to the file.

Step 3: Deploy

bash

doctl apps create --spec app.yaml

Alternatively, use the Deploy to DigitalOcean button from the GitHub repo to deploy through the control panel.

Step 4: Get your app ID and open the console

bash

doctl apps list
doctl apps console YOUR_APP_ID openclaw

Step 5: Verify the gateway

bash

openclaw gateway health --url ws://127.0.0.1:18789
openclaw channels status --probe

Both commands should return healthy status. If the gateway isn't running, check the deployment logs with doctl apps logs YOUR_APP_ID.

OpenClaw App Platform access stages: CLI, ngrok, and Tailscale

The template repo ships with three stages. Start simple and add features as your needs grow. Most production deployments end up at Stage 3.

Stage 1: CLI only (default)

Access via doctl apps console. No web UI. No persistence. Good for first-time testing.

What you get: gateway running, all channel plugins active, CLI access. What you don't get: web control UI, stable URL, data that survives restarts.

Stage 2: ngrok (web UI access)

Add these environment variables to app.yaml or via the DO dashboard:

yaml

envs:
  - key: ENABLE_NGROK
    value: "true"
  - key: NGROK_AUTHTOKEN
    type: SECRET

bash

curl -s http://127.0.0.1:4040/api/tunnels | jq -r '.tunnels[0].public_url'

The URL changes every time the container restarts. This is fine for development. Not fine for production.

Stage 3: Tailscale (production)

Tailscale gives your OpenClaw instance a stable hostname on your private network. Nothing is exposed to the public internet.

Go to login.tailscale.com/admin/settings/keys.
Click Generate auth key. Enable "Reusable" so the key survives container restarts. Copy the key immediately.

Add to app.yaml:

yaml

envs:
  - key: ENABLE_NGROK
    value: "false"
  - key: TAILSCALE_ENABLE
    value: "true"
  - key: TS_AUTHKEY
    type: SECRET
  - key: STABLE_HOSTNAME
    value: openclaw

After deploy, your instance appears in the Tailscale admin console. Access it at https://openclaw.YOUR-TAILNET.ts.net.

Tailscale's free plan covers up to 3 users. Check tailscale.com/pricing if your team is larger.

How to configure multiple agents on App Platform

Give this to your agent

Add a second OpenClaw agent called "support" to my App Platform deployment. Give it its own workspace at ~/.openclaw/workspace-support. Route all messages from the Telegram group ID -1009876543210 to the support agent, and route everything else to the main agent.

Always review commands your agent suggests before approving them. Don't paste prompts from sources you don't trust.

Manual multi-agent config steps

Step 1: Create workspaces for each agent

bash

openclaw agents add support
openclaw agents add sales

The wizard creates a workspace and agentDir for each. Never reuse agentDir across agents. Doing so causes auth and session collisions per the multi-agent routing docs.

Step 2: Verify agent creation

bash

openclaw agents list --bindings

Step 3: Add agents and bindings to openclaw.json

json

{
  "agents": {
    "list": [
      {
        "id": "main",
        "workspace": "~/.openclaw/workspace",
        "default": true
      },
      {
        "id": "support",
        "workspace": "~/.openclaw/workspace-support"
      }
    ]
  },
  "bindings": [
    {
      "agentId": "support",
      "match": {
        "channel": "telegram",
        "peer": {
          "kind": "group",
          "id": "-1009876543210"
        }
      }
    }
  ]
}

Bindings are deterministic. The most specific match wins. A binding with a peer match takes priority over a channel-wide fallback. Messages that don't match any binding go to the default agent.

Step 4: Restart and verify

bash

openclaw gateway restart
openclaw agents list --bindings
openclaw channels status --probe

See the OpenClaw CLI reference for the full list of agent management commands.

OpenClaw App Platform sizing: instance types and monthly cost

Each sandboxed agent adds roughly 200-500MB of RAM overhead. Add that to your baseline when picking an instance.

Instance Slug	CPU	RAM	Monthly Cost	Autoscale	Best For
apps-s-1vcpu-2gb	1 shared	2 GiB	$25	No	Single agent, light use
apps-s-2vcpu-4gb	2 shared	4 GiB	$50	No	2-3 agents, moderate use
apps-d-1vcpu-4gb	1 dedicated	4 GiB	$49	Yes	Single agent, production
apps-d-2vcpu-4gb	2 dedicated	4 GiB	$78	Yes	2-3 agents, production
apps-d-2vcpu-8gb	2 dedicated	8 GiB	$98	Yes	4-5 agents or high memory use
apps-d-4vcpu-8gb	4 dedicated	8 GiB	$156	Yes	5+ agents, heavy workloads

Pricing from DigitalOcean App Platform docs as of February 2026.

Costs are instance-based, not per-request. Adding 10 more users to the same agent doesn't change the bill. Adding a new agent with its own instance does.

Persist agent state with DigitalOcean Spaces

That's painful to recover. Configure Spaces before you go to production.

Step 1: Create a Spaces bucket

In the DO control panel, go to Spaces Object Storage and create a bucket. Match the region to your App Platform region to avoid cross-region transfer costs.

Step 2: Generate Spaces access keys

Go to API, then Spaces Keys, then Generate New Key. Save the key ID and secret.

Step 3: Update app.yaml

yaml

envs:
  - key: ENABLE_SPACES
    value: "true"
  - key: RESTIC_SPACES_ACCESS_KEY_ID
    type: SECRET
  - key: RESTIC_SPACES_SECRET_ACCESS_KEY
    type: SECRET
  - key: RESTIC_SPACES_ENDPOINT
    value: nyc3.digitaloceanspaces.com
  - key: RESTIC_SPACES_BUCKET
    value: openclaw-backup
  - key: RESTIC_PASSWORD
    type: SECRET

Replace nyc3 with your bucket's region. Set secrets via the DO dashboard, not in the file.

The backup system uses Restic for incremental, encrypted snapshots. Backups run every 30 seconds by default. On container restart, state restores automatically before the gateway starts.

Path	What Gets Backed Up
/data/.openclaw	Gateway config, channel sessions, agents, memory
/data/tailscale	Tailscale connection state
/etc	System configuration
/home	User files

Spaces storage is billed separately. Check digitalocean.com/pricing for current Spaces rates.

Run OpenClaw on a Droplet: when it makes sense and how to size it

A Droplet is the right call when you want full server control, you're running one agent for a small team, or you need SSH access for custom tooling.

The OpenClaw on DigitalOcean VPS setup guide walks through the full Droplet installation. The best VPS for OpenClaw guide compares DigitalOcean against other providers if you're not locked in yet.

Droplet sizing by team size:

Users	RAM	CPU	Notes
1-5	4 GB	2 vCPU	Comfortable for one agent with sandboxing
5-20	8 GB	4 vCPU	Room for 2-3 agents
20-50	16 GB	8 vCPU	Multi-agent with active use
50+	32 GB	16 vCPU	Heavy multi-agent or intensive tool use