AI-Native -- No Other Platform Has This

An AI that
understands and acts on
your infrastructure.

103 tools that read your logs, restart your apps, debug in a sandbox, and deploy new versions. Three safety layers so nothing destructive happens without your approval.

Get Started Built with Claude
103 AI tools
|
MCP server
|
3 safety layers
|
3 Claude models

Open your sh0 dashboard and navigate to /ai. All 103 tools available instantly.

Full setup guide  ·  Get AI credits  ·  103 tools reference

Three Claude models. One interface.

All models available to all plans. Pay per use from your wallet. Switch models with a single click.

Claude Haiku
FAST

Best for quick questions, CLI help, and simple debugging. Responds in under a second.

Input$1.20/MTok
Output$6.00/MTok
Per conversation~$0.002
Claude Sonnet
RECOMMENDED

The sweet spot. Code review, deployment debugging, infrastructure planning. Smart and fast.

Input$3.60/MTok
Output$18.00/MTok
Per conversation~$0.005
Claude Opus
POWERFUL

Maximum intelligence. Complex architecture decisions, deep debugging, multi-step reasoning.

Input$6.00/MTok
Output$30.00/MTok
Per conversation~$0.009

What the AI can do

Built specifically for server management and deployment workflows.

Debug failed deployments
Analyze logs, identify root causes, suggest fixes -- in seconds
Review Dockerfiles
Spot issues, suggest multi-stage builds, optimize image size
Write sh0.yaml configs
Generate infrastructure-as-code from natural language
Monitor & troubleshoot
Check server health, diagnose networking, fix DNS issues
Configure domains & SSL
Set up custom domains, wildcard certs, reverse proxy rules
Explain sh0 features
Get instant answers about any sh0 feature or API endpoint
Restart apps & trigger backups
The AI doesn't just advise -- it can restart containers, trigger backups, scale replicas, and deploy new versions
MCP server built-in
Connect Claude Desktop, Cursor, or any MCP-compatible client. 103 tools auto-discovered via Model Context Protocol
Safety gates on every action
Scoped API keys, risk classification, and confirmation tokens. Destructive operations require explicit approval
AI Sandbox -- always-on dev environment
A global sandbox starts with your server -- no apps needed. Clone repos, install packages, build and run apps. Root access, 1 GB RAM, 2 CPUs.

Not a chatbot. An AI ops engineer.

sh0's AI doesn't just give advice. It calls 20 real tools that read your infrastructure and take action.

Tool execution timeline

My app returns 502. Fix it.
Tools executed
get_server_status get_app_details get_deployment_logs
Found it. Your app api-prod crashed 3 minutes ago -- the latest deploy has an OOM error. The container exceeded its 512 MB memory limit. I'll redeploy with the previous working image.
Tools executed
deploy_app
Done. Rolled back api-prod to the previous image. Health check passed, app is returning 200. I recommend increasing the memory limit to 1024 MB and investigating the memory leak in the latest version.
Read Tools 8
  • list_apps
  • get_app_details
  • get_deployment_logs
  • get_server_status
  • list_databases
  • list_cron_jobs
  • list_backups
  • get_app_metrics
Write Tools 5
  • restart_app
  • deploy_app
  • scale_app
  • trigger_backup
  • trigger_cron_job
Destructive Tools 2
  • delete_app
  • delete_database
Requires confirmation token
Meta Tools 2
  • confirm_action
  • generate_config_file
Sandbox Tools 5
  • sandbox_exec_command
  • sandbox_read_file
  • sandbox_list_processes
  • sandbox_check_connectivity
  • sandbox_status
Global sandbox always available, per-app sandbox optional

Works with Claude Desktop, Cursor, and any MCP client

sh0 includes a full MCP server (Model Context Protocol) with Streamable HTTP transport. AI clients connect to your server, discover all 103 tools automatically, and manage your infrastructure via natural language -- no dashboard required.

Built-in

Dashboard Chat

Built into your sh0 dashboard. No setup needed -- open /ai and start chatting with full tool access.

MCP

Claude Desktop

Connect Claude Desktop to your sh0 server via MCP. Manage infrastructure from your desktop without opening the dashboard.

Universal

Any MCP Client

Cursor, custom agents, or any MCP-compatible client. 103 tools auto-discovered via Streamable HTTP transport.

Connect in 30 seconds

Replace your-server.com with your sh0 domain and sh0_your_api_key with an API key from Settings.

Claude Code / Desktop
.mcp.json or claude_desktop_config.json
{
  "mcpServers": {
    "sh0": {
      "type": "url",
      "url": "https://your-server.com/mcp",
      "headers": {
        "Authorization": "Bearer sh0_your_api_key"
      }
    }
  }
}
Cursor / Any MCP Client
.cursor/mcp.json
{
  "mcpServers": {
    "sh0": {
      "url": "https://your-server.com/mcp",
      "headers": {
        "Authorization": "Bearer sh0_your_api_key"
      }
    }
  }
}

Your MCP endpoint is always at https://your-sh0-domain/mcp -- 103 tools auto-discovered via Streamable HTTP transport.

Three layers of protection

The AI has real power over your infrastructure. That power comes with real safeguards.

Scoped API Keys

Three permission levels: read (view only), standard (read + write), and admin (full access including destructive operations). Each key has a defined scope.

Risk Classification

Every tool is classified by risk level: read (safe), write (modifies state), or destructive (irreversible). The AI announces the risk level before executing.

Confirmation Tokens

Destructive operations generate a single-use confirmation token with a 5-minute TTL, bound to the requesting user. The AI cannot bypass this step.

Every write operation is audit-logged with full traceability

How it works

01

Purchase AI credits at sh0.dev

Buy a credit pack ($5 to $100). Credits never expire. All three Claude models are available to every plan.

02

Create an AI API key

Go to sh0.dev/account/ai and generate a key. It starts with sh0_ai_ and is shown once -- copy it immediately.

03

Paste the key in your dashboard Settings

Open your sh0 dashboard, go to Settings, paste the key in the AI section. It is stored locally in your browser.

04

Open /ai and start chatting

The AI knows your server context -- apps, versions, metrics. Ask anything about deployments, Docker, networking, or code.

Real conversation example

My Django app deploy failed. The logs say "relation does not exist".
That's a missing database migration. Your deploy ran the app before applying the schema changes. Add a pre-deploy hook to run migrations automatically:
sh0 hooks set my-django-app \
  --pre-deploy "python manage.py migrate --noinput"
This will run migrations inside the container before the new version goes live. If the migration fails, the deploy will be aborted and the old version stays running.

Built for developers

Powered exclusively by Claude

The same AI that built sh0 itself. Claude knows server infrastructure, Docker, and deployment pipelines better than any other model.

Knows your server context

The AI automatically sees your server version, running apps, metrics, and plan. No need to explain your setup every time.

Secure by design

Your API key is stored locally in your browser. Requests are proxied through our encrypted gateway. We never see your conversations.

Copy-paste-ready commands

Every answer includes actionable commands you can run immediately. No vague suggestions -- concrete solutions.

3 models, one interface

Switch between Haiku (fast), Sonnet (smart), and Opus (powerful) with a single click. Use the right model for the job.

Conversation history

All conversations are saved locally. Search, favorite, and organize past chats. Pick up where you left off.

Without sh0 AI

  • xGoogle error messages for hours
  • xCopy-paste from Stack Overflow
  • xRead documentation page by page
  • xTrial and error with Docker configs
  • xContext-switch between tools
  • xDebug deployments blindly
  • xNo AI on any other hosting platform

With sh0 AI

  • Get the fix in one message
  • Copy-paste-ready commands
  • AI knows your server context
  • Dockerfile review in seconds
  • Everything inside the dashboard
  • Powered by Claude -- the best
  • 103 tools that read and act on your infrastructure
  • MCP server for Claude Desktop & Cursor

The roadmap

What's coming next for the sh0 AI assistant.

File Generation

Generate XLSX, PDF, and DOCX files directly. Export server reports, audit logs, and infrastructure documentation.

RAG Pipeline

Upload your own documentation, runbooks, and internal guides. The AI will reference them when answering questions.

Background Jobs

Long-running AI tasks that survive disconnects. Queue a complex analysis, close your browser, come back to results.

Questions & answers

Which AI models are available? +
Three Claude models: Haiku (fast, cheap), Sonnet (balanced, recommended), and Opus (most powerful). All models are available to all plans -- Free, Pro, and Business.
How does pricing work? +
You purchase credit packs ($5 to $100) that are deducted per-use based on token consumption. Credits never expire. We charge Anthropic's list price plus a 20% service fee. A $5 pack gives you roughly 2,500 Haiku conversations.
Can I use my own Anthropic or OpenRouter key? +
Yes -- Business plan users can bring their own key (BYOK). Enter your Anthropic (sk-ant-...) or OpenRouter (sk-or-...) key in your account settings. With BYOK, there is zero markup -- you pay the provider directly. Only Claude models are available regardless of provider.
Is my data secure? +
Your AI API key is stored only in your browser's localStorage. Chat messages are sent through our encrypted gateway to Anthropic. We log conversation metadata (token counts, costs) for billing but do not use your conversations to train models.
Does the AI have access to my server? +
Yes. The AI has 103 tools that can read and act on your infrastructure -- listing apps, checking logs, restarting containers, running commands in a sandbox, triggering backups, and more. Access is controlled by scoped API keys (read, standard, or admin) and destructive operations require a confirmation token. Every action is audit-logged with full traceability.
What can the AI help with? +
Debugging failed deployments, reviewing Dockerfiles and docker-compose.yml, writing sh0.yaml infrastructure code, configuring domains and SSL, setting up monitoring and alerts, troubleshooting networking and DNS, restarting apps, triggering backups, scaling replicas, deploying new versions, and explaining any sh0 feature or API endpoint.
What is the MCP server? +
sh0 includes a built-in MCP (Model Context Protocol) server using Streamable HTTP transport. MCP is an open standard that allows AI clients to discover and call tools on remote servers. You can connect Claude Desktop, Cursor, or any MCP-compatible client to your sh0 server and manage your infrastructure via natural language -- without opening the dashboard.
What are confirmation tokens? +
When the AI attempts a destructive operation (like deleting an app or database), it cannot execute immediately. Instead, the system generates a single-use confirmation token with a 5-minute TTL, bound to the requesting user. The AI presents the token and asks for your explicit approval. Only after you confirm does the operation execute. This prevents accidental or unauthorized destructive actions.
Can the AI delete my apps? +
Only if you use an admin-scope API key AND explicitly confirm the operation. Destructive tools (delete_app, delete_database) require a confirmation token -- a single-use, time-limited, user-bound approval. The AI cannot bypass this. With a read or standard scope key, destructive tools are entirely unavailable.

The only hosting platform with a built-in AI ops engineer

103 tools. MCP server. 3 safety layers. Powered exclusively by Anthropic Claude. No other platform comes close.

Get sh0 Read the Docs