Deploy a fleet of specialized OpenClaw AI agents on AWS or Hetzner Cloud — managed entirely from your terminal.
Clawup provisions autonomous AI agents with persistent memory, role-specific behavior, and secure networking. Each agent runs in a Docker sandbox with its own identity — personality, skills, tools, and model preferences — connected over a Tailscale mesh VPN with no public port exposure.
You define what your agents do. Clawup handles where and how they run.
┌─────────────────────────────────────────────────────────────────────┐
│ AWS VPC / Hetzner Cloud │
│ │
│ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐ │
│ │ Agent A │ │ Agent B │ │ Agent C │ │
│ │ │ │ │ │ │ │
│ │ • OpenClaw │ │ • OpenClaw │ │ • OpenClaw │ │
│ │ • Docker │ │ • Docker │ │ • Docker │ │
│ │ • Tailscale │ │ • Tailscale │ │ • Tailscale │ │
│ └──────┬───────┘ └──────┬───────┘ └──────┬───────┘ │
│ └──────────────────┼──────────────────┘ │
└────────────────────────────┼────────────────────────────────────────┘
│
┌────────────▼────────────┐
│ Tailscale Mesh VPN │
│ (Encrypted P2P) │
└────────────┬────────────┘
│
┌────────────▼────────────┐
│ Your Machine │
└─────────────────────────┘
Agents are defined by identities — composable, self-contained packages that include everything an agent needs to operate: personality, skills, model preferences, plugin configuration, and dependencies.
An identity is a directory with an identity.yaml manifest and workspace files:
my-identity/
├── identity.yaml # Manifest: model, plugins, deps, skills, template vars
├── SOUL.md # Personality, role, approach, communication style
├── IDENTITY.md # Name, role, emoji
├── HEARTBEAT.md # Periodic tasks and state machine logic
├── TOOLS.md # Tool reference (integrations, local env)
├── AGENTS.md # Multi-agent coordination instructions
├── BOOTSTRAP.md # First-run setup instructions
├── USER.md # Owner-specific info (templated)
└── skills/ # Bundled skills
└── my-skill/
└── SKILL.md
Identities can live in a Git repo, a monorepo subdirectory, or a local path. Point any agent at any identity:
# clawup.yaml
agents:
- name: agent-researcher
displayName: Atlas
role: researcher
identity: "https://github.com/your-org/your-identities#researcher"
identityVersion: "v1.0.0" # optional: pin to a tag or commit
volumeSize: 20The identity manifest declares the agent's defaults:
name: researcher
displayName: Atlas
role: researcher
emoji: telescope
description: Deep research, source analysis, report generation
volumeSize: 20
model: anthropic/claude-sonnet-4-5
codingAgent: claude-code
deps:
- brave-search
plugins:
- slack
pluginDefaults:
slack:
mode: socket
dm:
enabled: true
policy: open
skills:
- research-report
templateVars:
- OWNER_NAME
- TIMEZONE
- WORKING_HOURS
- USER_NOTESClawup ships with three built-in identities to get you started:
| Alias | Role | What It Does |
|---|---|---|
| Juno | PM | Breaks down tickets, researches requirements, plans & sequences work, tracks progress |
| Titus | Engineer | Picks up tickets, writes code via Claude Code, builds/tests, creates PRs |
| Scout | Tester | Reviews PRs, tests happy/sad/edge cases, files bugs, verifies fixes |
These are standard identities hosted in a Git repo — the same format as any custom identity you'd create.
See the examples/identity/ directory for a complete, minimal example (a "researcher" agent), and the Creating Identities guide for the full authoring reference covering:
- Identity structure and required files
identity.yamlfield reference- Workspace file conventions
- Skill authoring (private and public)
- Template variable substitution
- Available registries (deps, plugins, coding agents)
npm install -g clawupclawup initThe wizard walks you through:
- Prerequisites check — verifies Pulumi, Node.js, cloud provider CLI, and Tailscale
- Cloud provider — AWS or Hetzner Cloud
- Region & instance type — with cost estimates shown inline
- Secrets — Anthropic API key, Tailscale auth key (with instructions for each)
- Agent selection — pick from built-in identities, point to a Git repo or local directory, or mix both
- Optional integrations — Slack, Linear, GitHub per agent
- Review & confirm — see full config and estimated monthly cost
This generates a clawup.yaml manifest and sets all Pulumi config values automatically.
clawup deployWait 3-5 minutes for cloud-init to complete, then:
clawup validateclawup ssh <agent-name> # SSH by name, role, or aliasRun clawup --help for the full list.
| Command | Description |
|---|---|
clawup init |
Interactive setup wizard |
clawup deploy |
Deploy agents (pulumi up under the hood) |
clawup deploy -y |
Deploy without confirmation prompt |
clawup status |
Show agent statuses and outputs |
clawup status --json |
Status in JSON format |
clawup ssh <agent> |
SSH to an agent by name, role, or alias |
clawup ssh <agent> '<cmd>' |
Run a command on an agent remotely |
clawup validate |
Health check all agents via Tailscale |
clawup destroy |
Tear down all resources (with confirmation) |
clawup redeploy |
Update agents in-place (pulumi up --refresh) |
clawup redeploy -y |
Redeploy without confirmation prompt |
clawup destroy -y |
Tear down without confirmation |
clawup list |
List saved configurations |
clawup config show |
Display current config |
clawup config show --json |
Config in JSON format |
clawup config set <key> <value> |
Update a config value |
clawup config set <key> <value> -a <agent> |
Update a per-agent config value |
clawup config migrate |
Migrate old plugin config files into manifest |
clawup secrets set <key> <value> |
Set a Pulumi secret (e.g. API keys) |
clawup secrets list |
Show which secrets are configured (redacted) |
clawup push |
Push workspace files, skills, and config to running agents |
clawup webhooks setup |
Configure Linear webhooks for deployed agents |
clawup update |
Update clawup CLI to the latest version |
Agent resolution is flexible — all of these target the same agent:
clawup ssh juno # by alias
clawup ssh pm # by role
clawup ssh agent-pm # by resource name| Feature | AWS | Hetzner Cloud |
|---|---|---|
| 3x Agents (monthly) | ~$110-120 | ~$18-22 |
| Instance Type | t3.medium (2 vCPU, 4GB) | CX22 (2 vCPU, 4GB) |
| Storage | ~$2.40/month per 30GB | Included |
| Data Transfer | ~$5-10/month | 20TB included |
| Regions | Global (25+) | EU & US (5 locations) |
| Setup Complexity | Moderate (VPC, IAM) | Simple (API token) |
Use Hetzner for development and cost savings (~80% cheaper). Use AWS for production or global reach.
Each agent gets:
- Cloud instance (EC2 or Hetzner server) with Ubuntu 24.04 LTS
- Docker (for OpenClaw sandbox)
- Node.js v22, OpenClaw CLI, coding agent CLI (from registry), GitHub CLI
- Tailscale VPN (encrypted mesh, no public ports)
- Workspace files and skills injected from its identity
- AI model configured per-identity (with fallback support)
- Plugins and deps installed per-identity
All agents share a single VPC/network for cost optimization.
You need the following installed on your local machine before running clawup init. The init wizard checks for these and will tell you what's missing.
| Dependency | Why | Install |
|---|---|---|
| Node.js 18+ | Runtime for CLI and Pulumi program | nodejs.org |
| Pulumi CLI | Infrastructure provisioning | pulumi.com/docs/iac/download-install |
| Pulumi Account | State management and encrypted secrets | app.pulumi.com/signup |
| Tailscale | Secure mesh VPN to reach your agents | tailscale.com/download |
Pick one depending on where you want to deploy:
| Provider | Dependency | Install |
|---|---|---|
| AWS | AWS CLI (configured with credentials) | aws.amazon.com/cli — then run aws configure |
| Hetzner | API token with Read & Write permissions | console.hetzner.cloud → Project → Security → API Tokens |
Tailscale requires a few one-time setup steps:
- Create an account
- Enable HTTPS certificates (required for OpenClaw web UI)
- Generate a reusable auth key with tags
- Note your tailnet DNS name (e.g.,
tail12345.ts.net)
These are provisioned on the cloud instances via cloud-init — you do not need them locally:
- Docker, Node.js v22, OpenClaw CLI, coding agent CLI (e.g., Claude Code)
- Tailscale (agent-side)
- GitHub CLI, Brave Search, and other deps (per-identity)
- OpenClaw plugins: Linear, Slack (per-identity)
| Key | Required | Where to Get |
|---|---|---|
| Anthropic Credentials | Yes | API Key or OAuth token (claude setup-token) |
| Tailscale Auth Key | Yes | Tailscale Admin (reusable, with tags) |
| Slack Bot Token | No | Slack API — per agent |
| Linear API Token | No | Linear Settings — per agent |
| GitHub Token | No | GitHub Settings — per agent |
Two authentication methods are supported:
| Method | Token Format | Best For |
|---|---|---|
| API Key | sk-ant-api03-... |
Pay-as-you-go API usage |
| OAuth Token | sk-ant-oat01-... |
Pro/Max subscription (flat rate) |
The system auto-detects which type you provide and sets the correct environment variable.
For in-place updates that preserve Tailscale devices and existing infrastructure:
clawup redeployThis runs pulumi up --refresh to sync cloud state and apply changes. If the stack doesn't exist yet, it falls back to a fresh deploy automatically.
For a clean rebuild (when in-place update can't recover):
clawup destroy -y && clawup deploy -yView your current configuration without opening the manifest file:
clawup config show # Human-readable summary
clawup config show --json # Full JSON outputModify config values with validation (no need to re-run init):
clawup config set region us-west-2
clawup config set instanceType t3.large
clawup config set instanceType cx32 -a atlas # Per-agent override
clawup config set volumeSize 50 -a atlas # Per-agent volumeRun clawup redeploy after changing config to apply.
Generated by clawup init. This manifest drives the entire deployment:
stackName: dev
provider: aws
region: us-east-1
instanceType: t3.medium
ownerName: Your Name
timezone: America/New_York
workingHours: 9am-6pm
agents:
- name: agent-pm
displayName: Juno
role: pm
identity: "https://github.com/your-org/army-identities#pm"
volumeSize: 30
plugins:
- openclaw-linear
- slack
deps:
- gh
- brave-search
- name: agent-researcher
displayName: Atlas
role: researcher
identity: "./my-identities/researcher"
volumeSize: 20Model, backup model, and coding agent are configured in the identity (not the manifest). The manifest defines which agents to deploy and where.
Secrets are stored encrypted in Pulumi config, set automatically by the init wizard. You can also manage them directly:
pulumi config set --secret anthropicApiKey sk-ant-xxxxx
pulumi config set --secret tailscaleAuthKey tskey-auth-xxxxx
pulumi config set tailnetDnsName tail12345.ts.netFor more advanced secret management, use Pulumi ESC. See esc/clawup-secrets.yaml.example for the full template.
clawup/
├── packages/
│ ├── core/ # @clawup/core — shared types, constants, registries
│ │ └── src/
│ │ ├── schemas/ # Zod schemas (source of truth for types)
│ │ ├── constants.ts
│ │ ├── identity.ts # Identity loader (Git repos, local paths)
│ │ ├── plugin-registry.ts
│ │ ├── coding-agent-registry.ts
│ │ ├── dep-registry.ts
│ │ └── skills.ts
│ ├── cli/ # clawup CLI (published npm package)
│ │ ├── bin.ts # Entry point (Commander.js)
│ │ ├── commands/ # init, deploy, redeploy, status, ssh, validate, destroy, config, list, push, secrets, webhooks, update
│ │ ├── tools/ # Tool implementations (adapter-based)
│ │ ├── lib/ # CLI-only: config, pulumi, ui, tailscale, exec
│ │ └── adapters/ # Runtime adapters (CLI vs API)
│ ├── pulumi/ # @clawup/pulumi — infrastructure as code
│ │ └── src/
│ │ ├── components/
│ │ │ ├── openclaw-agent.ts # AWS EC2 agent component
│ │ │ ├── hetzner-agent.ts # Hetzner Cloud agent component
│ │ │ ├── cloud-init.ts # Cloud-init script generation
│ │ │ └── config-generator.ts # OpenClaw config builder
│ │ ├── shared-vpc.ts
│ │ └── index.ts # Main Pulumi stack program
│ └── web/ # Next.js dashboard (clawup-web)
├── identities/ # Built-in identity stubs (point to external repos)
├── examples/ # Example identities for reference
│ └── identity/ # Complete "researcher" identity example
├── docs/ # Documentation (Mintlify site + guides)
├── esc/ # Pulumi ESC secret templates
├── scripts/ # Shell script helpers
├── Pulumi.yaml # Pulumi project config (points to packages/pulumi/)
└── pnpm-workspace.yaml # Monorepo workspace config
- All agent ports bind to
127.0.0.1— access is via Tailscale only - No public port exposure; Tailscale Serve proxies traffic
- Token-based gateway authentication
- Secrets encrypted via Pulumi config
- Cloud-init scripts use environment variable interpolation
- SSH available as fallback for debugging
- Wait 3-5 minutes for cloud-init to complete
- Check logs:
clawup ssh <agent> 'sudo cat /var/log/cloud-init-output.log | tail -100' - Verify your Tailscale auth key is valid and reusable
clawup ssh <agent> 'openclaw gateway status'
clawup ssh <agent> 'journalctl -u openclaw -n 50'
clawup ssh <agent> 'openclaw gateway restart'- Check Tailscale is running locally:
tailscale status - Verify the agent appears in your tailnet
- Ensure you're using the correct tailnet DNS name
pulumi refresh # Refresh state from actual infrastructure
pulumi cancel # Force unlock if lockedFor contributing to Clawup itself:
git clone https://github.com/stepandel/clawup.git
cd clawup
pnpm install
pnpm build # Build all packages
pnpm test # Run all tests
# Individual packages
pnpm --filter @clawup/core build # Build core
pnpm --filter clawup build # Build CLI
pnpm --filter @clawup/pulumi build # Build Pulumi
pnpm --filter clawup-web dev # Web dev serverMIT