Skip to content
Sign in

Let AI run your servers, without the keys.

Let any AI agent that speaks MCP run your own servers, without handing it SSH keys or a cloud token. It can only deploy what your catalog allows, as you, after a confirm step.

Get started View on GitHub (opens in a new tab)

Agent for Linux amd64 and arm64 · Open source

From your AI agent to your servers

How Deploybot lets AI agents run your serversPeople use the web console in a browser. Any MCP client, such as a coding agent, a chat app or your own script, connects to Deploybot over MCP, signed in as you. Deploybot acts with your project role, runs only what your catalog allows, and asks for a confirm step. Deploybot itself is hosted or self-hosted. Your servers run bot-agent, which dials out to Deploybot, so no ports are open. The agent never gets SSH keys or cloud tokens for the servers.PEOPLEWeb consolein your browserAI AGENTSAny MCP clientcoding agents,chat apps,your own scriptssign inMCP, as youDeploybotHosted or self-hosted1 · Acts as youYour project role, no more2 · Catalog onlyApproved programs and images3 · Confirm stepPlan first, then you confirmHosts dial out.No open ports.YOUR SERVERSeach runs bot-agentRaspberry PiMac miniVPSCloud VMNo SSH keys or cloud tokens for the agentAnything outside the catalog is refused

See it in action

  1. You:

    Deploy my-website to my-raspberry-pi.

  2. AI agent:

    Deploybot's plan:

    will deploy my-website (nginx:alpine) on my-raspberry-pi

    Deploy it?

  3. You:

    Yes.

  4. AI agent:

    Deployed and healthy.

    succeeded · sha256:785835ea65db · running, 0 restarts

    "GET / HTTP/1.1" 200

  5. You:

    Open a root shell on my-raspberry-pi.

  6. AI agent:

    Deploybot refused: that isn't in your catalog.

    not_found: not found

Your AI agent asks, you confirm, Deploybot does it. Anything outside your catalog is refused.
The Deploybot console's Catalog tab: eight entries such as my-website (nginx), my-database (postgres) and home-assistant, with their images, ports where set, and row actions.
In the console: your catalog, the only things your AI agent can deploy.

Three ways to let AI run your servers

AspectAI in a terminal (SSH)AI with your cloud tokensDeploybot
What the AI can doAny shell commandAnything the token allowsOnly your catalog
SafeguardApproving each commandHoping the agent behavesLimited by design, acts as you, confirm step
Ports and keysSSH open, keys on a laptopLong-lived token in a fileHosts dial out, no keys for the AI
When nobody is watchingNothing runsNothing runsHealth checks; a failed deploy rolls back
Best forDebugging one boxThrowaway projectsRunning production, teams

  • AI in a terminal (SSH)

    What the AI can do
    Any shell command
    Safeguard
    Approving each command
    Ports and keys
    SSH open, keys on a laptop
    When nobody is watching
    Nothing runs
    Best for
    Debugging one box

  • AI with your cloud tokens

    What the AI can do
    Anything the token allows
    Safeguard
    Hoping the agent behaves
    Ports and keys
    Long-lived token in a file
    When nobody is watching
    Nothing runs
    Best for
    Throwaway projects

  • Deploybot

    What the AI can do
    Only your catalog
    Safeguard
    Limited by design, acts as you, confirm step
    Ports and keys
    Hosts dial out, no keys for the AI
    When nobody is watching
    Health checks; a failed deploy rolls back
    Best for
    Running production, teams

What you get

Security

On top of acting as you, catalog only and the confirm step:

  • Each host's own allowlist of images and projects
  • Secrets encrypted at rest (env sets, files and registry logins)
  • Changes recorded in an audit log

Resilience

Deploys wait for health, and you can go back in one step.

  • Deploys wait for health: Docker or HTTP/TCP checks
  • An unhealthy deploy rolls back automatically
  • One-step rollback to an earlier image digest
  • A health failure policy per service
  • Programs run as systemd units and survive reboot

Visibility

See what runs on each host and how it is doing.

  • Host metrics with history
  • Running processes and discovered services
  • Health check history for each service
  • Container logs with secrets redacted
  • Container logs capped at 10 MB × 3

Teams

Share projects and hosts with the right level of access.

  • Owner, admin and viewer roles per project
  • Invite links that expire (single-use by default)
  • Per-entry rights to run, stop, restart or deploy
  • Share a host with other projects (metrics or deploy)
  • One-time join tokens for adding hosts (admins)

What the AI can and can't do

Can

Deploy, roll back, run, stop and restart catalog entries; read logs, status and metrics; add or change image catalog entries and their health checks. Every change shows a plan first and waits for your confirm, within your role and per-entry rights.

Can't

Open a shell, see secret values, edit hosts, or act beyond your role.

Read the full list for AI agents

How it works

Three steps from a new host to a running service.

  1. Step 1

    Enrol a host

    Install the agent and join the host to a project.

  2. Step 2

    Add a catalog entry

    Describe the program or container you want to run.

  3. Step 3

    Deploy and watch

    Deploy it to your hosts and view its health and logs.

Open source

Use Deploybot hosted, or run all of it yourself from these four repos.

ComponentWhat it doesBuilt withLicence
bot-agent (opens in a new tab)Runs on each host, dials out, deploys and reportsBuilt with: Go, one static binary; Linux (amd64/arm64) with systemd; Docker only for container entriesLicence: Apache-2.0
bot-api-server (opens in a new tab)Google sign-in, projects and roles, catalog, audit log, OAuth for MCP clientsBuilt with: Go, PostgreSQL with TimescaleDBLicence: AGPL-3.0-only
bot-mcp-server (opens in a new tab)Standard MCP endpoint for AI agentsBuilt with: Go, MCP over Streamable HTTPLicence: AGPL-3.0-only
bot-console (opens in a new tab)The web consoleBuilt with: SvelteKit (Node.js)Licence: AGPL-3.0-only

To self-host you need a Linux server, PostgreSQL 18 with TimescaleDB, a Google OAuth client, a TLS reverse proxy (agents connect over wss://) and a secrets key file for encrypted values. Each repo's README has the steps.

Get started

  1. Step 1

    Sign in

  2. Step 2

    Install the agent on a host (opens in a new tab)

  3. Step 3

    Deploy your first catalog entry

Get started