Your AI Platform Engineer.
One command. One VPS. You get your own cloud
with an AI agent that deploys, monitors, and operates everything.
Powered by Claude Code — running inside a Docker container, operating your infrastructure via natural language.
What you get · Quickstart · Workflows · Talk to it · Architecture · Contributing
demo-pleng-compressed.mp4
curl -fsSL https://raw.githubusercontent.com/mutonby/pleng/main/install.sh | sudo bashThat's it. Fresh Ubuntu VPS → Docker, Pleng, SSL, Telegram bot, AI agent — all configured and running. The installer asks for your tokens interactively and handles everything.
Or if you prefer doing it manually:
git clone https://github.com/mutonby/pleng && cd pleng
cp .env.example .env # add your tokens
docker network create pleng_web
docker compose up -dInstall Pleng on any VPS and you get all of this out of the box:
| What you'd normally install separately | Pleng gives you |
|---|---|
| Coolify / Dokploy — deploy apps, reverse proxy, SSL | Deploy anything via Telegram. Traefik + Let's Encrypt automatic. |
| Uptime Kuma — monitoring + alerts | Health checks every 10 min. Auto-restart. Telegram alerts. |
| Datadog / Grafana — observability, metrics, logs | Agent inspects CPU, RAM, disk, Docker stats, Traefik errors, container logs — all via natural language. |
| PagerDuty / OpsGenie — intelligent alerting | AI-powered heartbeat: the agent reviews your system at increasing depth. Reports anomalies to Telegram. |
| Plausible / analytics — traffic monitoring | Built-in analytics from Traefik access logs. Pageviews, visitors, top pages, sources. Zero-install. |
| Restic / backup tools — automated backups | Daily automatic backups of Pleng state (DB + site configs). Keeps last 7 days. Telegram notification. For full data (app databases, uploads), use your VPS provider's snapshots. |
| OpenClaw / AI agent — an AI that does things for you | Claude Code agent that writes code, deploys, diagnoses, operates. |
Right now, if you want an AI agent (OpenClaw, Claude Code, Cursor) to deploy something, you have two bad options:
-
Give it
--skip-permissionson your machine. It can run any command. Deploy one app? Sure. But deploy 3? It starts conflicting ports, overwriting configs, breaking your other projects. And if it hallucinates a badrm -rf, your machine pays the price. -
Give it SSH access to a VPS. Same problem, worse consequences. The agent has root on a server. No isolation. No rollback. No way to know what it changed.
Both approaches work for a quick demo. Neither works when you have real projects running.
Isolation. The agent runs inside a sandboxed container. No Docker socket. No SSH. No root. It can only affect infrastructure through the pleng CLI, which talks to the platform API over HTTP. If the agent goes rogue, it can't break your server — it only has the tools you gave it.
Abstraction. The agent doesn't think about ports, Traefik configs, SSL certs, or Docker networks. It says pleng deploy and the platform handles everything. Deploy 10 apps — each gets its own subdomain, its own containers, its own compose. No conflicts.
Scalability. Each app is a separate Docker Compose project. They don't know about each other. You can have 20 staging sites running simultaneously, each with its own URL, and promote any of them to production with a custom domain whenever you want.
Safety. Production sites can't be accidentally deleted — remove keeps their files. The agent needs explicit --confirm yes to permanently destroy anything. Health monitoring auto-restarts crashed containers. Daily backups protect your state.
Coolify, Dokploy, Plausible — they're dashboards built for humans clicking buttons. They have no API that an AI agent can use. They don't understand natural language. They can't be operated programmatically.
Pleng is designed from the ground up for AI agents. Every operation is available through:
- Natural language (Telegram / terminal)
- REST API (for any agent or script)
- skill.md — a machine-readable instruction file that any AI agent can read and immediately know how to deploy, manage, and monitor your apps
This means you can connect any AI agent to your Pleng:
# OpenClaw, Claude Code, Cursor, Windsurf, any agent:
"read http://panel.YOUR-IP.sslip.io/skill.md and deploy my project"The agent reads skill.md, learns the API, authenticates with the API key, and starts deploying. No plugins. No integrations. One URL.
All self-hosted. All in 5 containers. All operated by talking to it.
Before Pleng, to self-host a few apps you'd need to:
- Install Docker, configure networking
- Set up Traefik or Nginx as reverse proxy
- Configure Let's Encrypt for SSL
- Deploy each app manually with docker compose
- Set up monitoring (Uptime Kuma, Grafana)
- Configure alerts (email, Slack, Telegram)
- Handle DNS, domains, subdomains
- Debug issues by reading logs manually
- Restart crashed services manually
With Pleng, you skip all of that. You talk to the agent and it handles everything.
You already have a project (built with Claude Code, Cursor, by hand, whatever). You just want it live.
You: "deploy github.com/user/my-app"
Pleng: Cloning... building... deploying...
Live at http://a3f2.89.141.205.249.sslip.io
Or send a docker-compose.yml via Telegram. Or use the API from your local Claude Code.
You don't have code. You describe what you want. The agent writes everything and deploys.
You: "build me a booking API with Postgres and a simple frontend"
Pleng: [writes Node.js API, Dockerfile, docker-compose with Postgres]
[deploys everything]
Live at http://fe01.89.141.205.249.sslip.io
Want changes? Just say so:
You: "add email field to the bookings endpoint"
Pleng: [edits code, redeploys]
Updated and live.
Your staging app is ready. Time for a real domain with HTTPS.
You: "put bookings.mydomain.com on my-app"
Pleng: [configures Traefik, Let's Encrypt generates SSL cert]
Live at https://bookings.mydomain.com
Point your domain's DNS A record to your VPS IP. Pleng handles the rest.
The agent monitors your sites. If something goes down, it tells you and tries to fix it.
Pleng (Telegram alert): 🔴 my-app is DOWN. Connection refused.
Pleng: 🔄 Auto-restarted my-app.
Pleng: 🟢 my-app is back UP.
Or you ask:
You: "why is my-app slow?"
Pleng: [reads Docker logs]
Container OOMKilled — out of memory.
Want me to increase the limit?
Pleng includes a heartbeat system where the AI agent periodically reviews your entire server. It's configured in a single file (heartbeat.md) with three check levels:
| Level | Every | What it does |
|---|---|---|
| Quick | 30 min | Container status, RAM, disk, load — a quick glance. Silent if OK. |
| Deep | 60 min | Reads logs from every container, checks Traefik errors, analyzes resource usage. |
| Full | 120 min | Complete system audit. Everything above plus trends and recommendations. |
The heartbeat runs in your same Telegram conversation — the agent has full context of your chat, your deploys, and the system state. It's not a dumb uptime ping; it's an AI reading your logs and telling you what's wrong.
⚡ Heartbeat quick
Container pleng-my-app-web-1 is restarting in a loop.
Last log: "Error: ECONNREFUSED 127.0.0.1:5432"
→ The Postgres container is down. Run: pleng restart my-app
Edit from Telegram. The heartbeat.md file lives on a persistent volume. Tell the agent "edit heartbeat.md and add a MongoDB check to the full heartbeat" — it does it, and the change survives redeploys.
You: "logs for my-app" → Docker logs
You: "restart my-app" → Done
You: "stop the demo" → Stopped
You: "list my sites" → All sites with URLs and status
You: "redeploy my-app" → Rebuild + restart
You: "how's the server doing?" → Full system status
Working with Claude Code, Cursor, Windsurf, OpenClaw, or any AI agent? Just paste this:
Read http://panel.YOUR-IP.sslip.io/skill.md — API key: pleng_xxx. Deploy this project.
Your agent reads the skill.md, learns the API, and deploys your code to your server. No plugins. No integrations. One message.
The skill.md is auto-generated and always up to date with your server's IP, API URL, and all available operations. The agent reads it once and knows how to:
- Deploy from git or upload
- Check status and logs
- Stop, restart, redeploy
- Promote to production with SSL
One URL. That's the entire integration.
Same agent, four interfaces:
| Interface | Best for |
|---|---|
| Telegram | Quick commands from anywhere. Deploys, status checks, alerts. |
| Terminal | Deep work on the VPS. Full Claude Code experience. make chat |
| Dashboard | Visual overview. Sites, logs, status. http://panel.IP.sslip.io |
| skill.md | External AI agents. Your Claude Code at home deploys to your server. |
- A VPS with Docker installed (any provider — Hetzner, DigitalOcean, Contabo)
- 2GB+ RAM
- Ports 80 and 443 open
- A Telegram bot token (free, from @BotFather)
git clone https://github.com/mutonby/pleng
cd pleng
cp .env.example .envEdit .env:
TELEGRAM_BOT_TOKEN=123456:ABC... # From @BotFather
TELEGRAM_CHAT_ID=123456789 # Your chat ID
PUBLIC_IP=89.141.205.249 # Your VPS IP (curl ifconfig.me)
ACME_EMAIL=you@example.com # For SSL certificatesFor Claude Code auth, either:
- OAuth (default): run
docker exec -it pleng-agent-1 sudo -u claude env HOME=/home/claude claude /loginafter first start - API key: add
CLAUDE_AUTH_MODE=api_keyandANTHROPIC_API_KEY=sk-ant-...to.env
Start:
docker compose up -d- Open
http://panel.YOUR-IP.sslip.io— you should see the dashboard - Message your bot on Telegram: "hello"
- If using OAuth, do the login step above
- Try: "build me a simple landing page and deploy it"
- Check the staging URL it gives you
┌──────────────────────────────────────────────────────┐
│ Your VPS │
│ │
│ ┌──────────┐ ┌──────────────┐ ┌────────────────┐ │
│ │ Traefik │ │ Platform API │ │ Agent │ │
│ │ (proxy) │ │ Docker sock │ │ (Claude Code) │ │
│ │ SSL/HTTP │ │ SQLite │ │ pleng CLI │ │
│ │ sslip.io │ │ Health mon. │ │ /projects vol │ │
│ └──────────┘ │ Backups │ └────────────────┘ │
│ ▲ │ Analytics │ │ │
│ │ └──────────────┘ │ │
│ │ ▲ │ │
│ │ │ HTTP (internal network) │
│ │ │ │ │
│ ┌──────────┐ │ ┌───────────────┐ │
│ │ Telegram │─────────┘ │ Dashboard │ │
│ │ Bot │ │ (React) │ │
│ └──────────┘ └───────────────┘ │
│ │
│ + your deployed apps (each in its own containers) │
└──────────────────────────────────────────────────────┘
5 containers. One docker compose up.
| Container | What it does |
|---|---|
| traefik | Reverse proxy. Free staging URLs via sslip.io. HTTPS via Let's Encrypt for production. |
| platform-api | Orchestrates Docker. REST API. SQLite state. Health monitoring. Backups. Analytics (Traefik logs). |
| agent | Claude Code with pleng CLI. Writes code, deploys, diagnoses. Isolated — no Docker socket. |
| telegram-bot | Bridges Telegram ↔ agent. Sends health alerts. |
| dashboard | React web panel. Sites, logs, status. Read-only — all operations via agent. |
- Agent is sandboxed. Claude Code runs inside a Docker container with no Docker socket, no host access, no sudo to the host. It can only affect infrastructure through the
plengCLI → platform-api HTTP calls. If the agent hallucinates or goes rogue, it can't break your server — it only has the tools you give it. - sslip.io for staging.
anything.YOUR-IP.sslip.ioresolves to your IP. No DNS config needed. Free. Instant. - Platform-api owns Docker. Single point of control for all container operations.
- SQLite, not Postgres. Zero extra containers. One file. Enough for single-VPS scale.
- Health monitoring built in. HTTP checks every 10 min + AI heartbeat at 3 depth levels. Auto-restart on failure. Telegram alerts.
- Editable config from Telegram. Both
CLAUDE.md(agent instructions) andheartbeat.md(monitoring config) live on persistent volumes. Edit them from Telegram — changes survive redeploys. pleng_webis an external network. Created once (by the installer or manually) and markedexternal: truein docker-compose.yml. User-deployed apps join this network so Traefik can route to them. Because it's external,docker compose downnever removes it — your deployed sites stay connected even when you restart Pleng itself.- Your docker-compose.yml is never modified. Pleng generates its own overlay file for Traefik labels.
Inside the agent, Claude Code has these tools:
# Deployment & management
pleng sites # List all sites
pleng deploy <path> --name <name> # Deploy a project
pleng deploy-git <url> --name <name> # Deploy from git
pleng redeploy <name> # Rebuild + restart
pleng logs <name> # Docker logs
pleng status <name> # Container info
pleng stop <name> # Stop
pleng restart <name> # Restart
pleng remove <name> # Remove
pleng promote <name> --domain <domain> # Production + SSL
# Observability & diagnostics
pleng system # CPU, RAM, disk, load, uptime
pleng docker-ps # All containers on the host
pleng docker-stats # CPU + RAM per container
pleng errors [--minutes 60] # Recent Traefik 5xx errors
pleng logs-summary # Recent errors from ALL sites
pleng health-report # Full system report (all of the above)| Coolify | Dokploy | Railway | Pleng | |
|---|---|---|---|---|
| Self-hosted | Yes | Yes | No | Yes |
| AI agent | No | No | No | Yes |
| Natural language ops | No | No | No | Yes |
| Build apps from description | No | No | No | Yes |
| Telegram control | No | No | No | Yes |
| Auto-restart + alerts | Plugin | No | No | Built-in |
| AI-powered monitoring | No | No | No | Yes — heartbeat.md |
| System observability | Dashboard | Dashboard | Dashboard | AI reads logs + metrics for you |
| skill.md for agents | No | No | No | Yes |
| Free staging URLs | No | No | Auto | Auto (sslip.io) |
| Setup | Complex | Medium | Cloud | docker compose up |
| Price | Free | Free | $5-20/mo | Free |
Coolify is a great dashboard for deploying apps. Pleng is an AI agent that operates your entire server. Different tools for different workflows. If you want to click buttons, use Coolify. If you want to talk to your infra, use Pleng.
| Variable | Description |
|---|---|
TELEGRAM_BOT_TOKEN |
From @BotFather |
TELEGRAM_CHAT_ID |
Your Telegram chat ID |
PUBLIC_IP |
Your VPS public IP (curl ifconfig.me) |
ACME_EMAIL |
Email for Let's Encrypt SSL certificates |
| Variable | Default | Description |
|---|---|---|
ANTHROPIC_API_KEY |
— | If using API key auth (instead of OAuth) |
CLAUDE_AUTH_MODE |
oauth | oauth or api_key |
MODEL_NAME |
claude-sonnet-4-20250514 | Claude model for the agent |
GITHUB_TOKEN |
— | For deploying from private repos |
WEB_UI_PASSWORD |
admin | Dashboard login password |
To push/pull code from GitHub, add a Personal Access Token:
- Go to github.com/settings/tokens → Generate new token (classic) → select
reposcope - SSH into your VPS and add it:
echo "GITHUB_TOKEN=ghp_your_token_here" >> /root/pleng/.env
cd /root/pleng && docker compose restart platform-apiNow you can use pleng push and pleng pull to sync code with GitHub. The token is only used by the platform API through deterministic CLI tools (pleng push/pleng pull) — the AI agent never has direct access to it.
Bot not responding?
- Check
docker compose logs telegram-bot— is it polling? - Verify
TELEGRAM_BOT_TOKENandTELEGRAM_CHAT_IDin.env - The bot only responds to the configured chat ID (security)
Deploy failed?
- Check
docker compose logs agentfor Claude Code errors - Check the site's build log in the dashboard
- OAuth not configured? Run the login command (see Quickstart)
HTTPS not working after promote?
- Verify your domain's DNS A record points to your VPS IP
- Check
docker compose logs traefikfor Let's Encrypt errors ACME_EMAILmust be a real email (not example.com)
network pleng_web was found but has incorrect label?
- The
pleng_webnetwork must exist before runningdocker compose up. Create it manually:docker network create pleng_web - If you already ran compose without it, stop everything (
docker compose down), remove the stale network (docker network rm pleng_web), create it fresh, and start again.
Agent says "starting up"?
- The agent waits for platform-api to be ready. Give it 30 seconds.
AGPL-3.0 — same as Coolify. Self-host freely. If you modify the code and offer it as a service, you must open-source your changes. See LICENSE.
One VPS. One command. Your own cloud with an AI that runs it.
