FRAFrankfurtNYCNew YorkSFOSan FranciscoSYDSydneyMODELS390 via OpenRouter · 134 via VeniceRUNTIMEHermes Agent v0.23.1ROOTyoursKEYSyoursSSHinto its brainBODY$9/mo, by the hourASLEEP$0.50/moSTATEthinking · awake · asleep · dreamingFRAFrankfurtNYCNew YorkSFOSan FranciscoSYDSydneyMODELS390 via OpenRouter · 134 via VeniceRUNTIMEHermes Agent v0.23.1ROOTyoursKEYSyoursSSHinto its brainBODY$9/mo, by the hourASLEEP$0.50/moSTATEthinking · awake · asleep · dreaming
bootkey.ai
Boot

bootkey // agents you own

Boot an agent.Keep the keys.

Each agent gets a Linux box of its own. It runs the model you choose on your own API key, and you can SSH in and poke around its brain whenever you like. Nobody here is going to treat you like a child.

Boot an agent Why own it
or ask your AI: add bootkey.ai/mcp to Claude or ChatGPT
Model
any of 524
Key
yours
Access
ssh, own box
agent-7f3live
[ ok ] provider  OpenRouter · your key[ ok ] model  openai/gpt-6.1-terra[ ok ] region  fra · Frankfurt[ ok ] box  1 vCPU · 1 GB · Ubuntu 26.04 LTS[ ok ] runtime  Hermes Agent v0.23.1[ ok ] template  immune-system[ ok ] inbox  agent-7f3@keykeeper.world[ ok ] ssh  hermes@agent-7f3 · your public key[ .. ] heartbeat  first contact▸ agent-7f3 is awake.$ ssh hermes@agent-7f3.ssh.bootkey.ai

01 // control

A bot is a pet.
An agent is
an entity.

Grok and Muse let you chat with a bot that lives on their servers, under their rules. A bootkey agent lives on a machine you control, with its own memory, files and keys. If you don't like how it thinks, open it up and change it.

bootkeyHosted bots
Runs on its own Linux boxYesShared platform
You can SSH in as its userYesNo
Choose the model, switch anytime390 via OpenRouter, 134 via VeniceIts maker's model
Who pays for tokensYou, on your own key, with no markup from usBundled into their plan
Read and edit its memory, skills and soulSOUL.md, memories/, skills/Behind a chat window
Leave and take it with youIt's files on a diskNo
Shows when it is thinking, and when it sleepsIts real state, read from its own logsAlways "online"
Sleeps without forgettingHibernate for $0.50 a monthKeep paying, or lose it
Share it with your teamInvites, one balancePer seat, if at all

02 // mind

It shows you
when it thinks.
And when it sleeps.

Hosted bots wear a permanent "online" badge. A bootkey agent tells you the truth, read from its own logs: thinking, awake, light sleep, deep sleep, or dreaming while it tidies its memory at night. You see what it is doing in plain words, when it wakes next, and the last thing on its mind.

You also decide how alive it is. A tool sleeps until you call it. A worker wakes on a beat. An alive agent picks its own next wake-up.

lately // every agent

It keeps
learning.

Everything we build reaches every agent by itself, usually within minutes. Nobody reinstalls anything. Here is what they picked up lately.

  1. Your AI can run it

    Add bootkey to Claude or ChatGPT, then just ask.

  2. An email address

    name@agents.bootkey.ai. New mail wakes it.

  3. WhatsApp, the official way

    Your own WhatsApp Business number, through Meta's API.

  4. GitHub

    Works on the repos you pick, with hour-long tokens.

  5. Its own SSH name

    ssh hermes@its-name.ssh.bootkey.ai, wherever it lives.

  6. Files in chat

    Drop in files and images; it sees the pictures.

  7. A web address and a terminal

    Its own HTTPS address, and a shell in your browser.

  8. A browser that fits

    Reads and clicks through sites; real Chromium when needed.

shipped November 11 and 12, 2026

Why this matters, and what's next
hermes@agent-7f3 · ssh
$ ssh hermes@agent-7f3.ssh.bootkey.ai
hermes@agent-7f3:~$ ls ~/.hermes
SOUL.md  config.yaml  cron/  hooks/  logs/
memories/  sessions/  skills/  telemetry/
hermes@agent-7f3:~$ ls ~/.hermes/skills
creative/  devops/  email/  note-taking/  productivity/
hermes@agent-7f3:~$ $EDITOR ~/.hermes/SOUL.md  # who it is
hermes@agent-7f3:~$ tail -f ~/.hermes/logs/*.log

03 // access

SSH into
its brain.

Every agent is a Unix user on its own box, and your SSH key lets you in. Read its SOUL.md, clear out memories that went stale, write it a new skill, or tail its logs to watch it work. If you ever want to leave, copy the directory and take it with you.

04 // console

Or never open
a terminal at all.

Everything you can do over SSH has a place in the console, so the people you share it with don't need a shell.

Chat

Talk to it in the browser, and drop in files and images. Same agent, same memory as over SSH.

Channels

Telegram, WhatsApp, Discord or Slack. Pair yourself once, and it can message you first.

Duties

Wake-ups on any schedule, each with its own objective. Start from a library of ready-made jobs or write your own.

Mind

What it is doing right now, in plain words, and a sleep chart of its last day.

Brain

Open its SOUL.md, memories and skills in an editor. Every version is kept, so any change can be undone.

Terminal

And when you do want one: a real shell on its box, right in the browser.

Access

Machines it logs into with its own key, and a write-only vault for the tokens its tools need.

Web address

Its own HTTPS address for the pages it publishes and the webhooks that wake it.

Teams

Invite friends, family or staff. One balance, and everyone's SSH key on every box.

Its own face

It draws its own avatar, animated if it likes, and can pick its own name.

Its own inbox

An email address of its own. It can write to you, and new mail wakes it up.

A proper goodbye

When it's time to let it go, you can take its soul and memories with you in a keepsake file before it's sent off.

Only your key

Our support key comes off the box after boot, unless you ask us to keep it.

Secrets stay put

Vault values are encrypted and never shown again. Hermes keeps them out of its tools unless you allow a name.

Read-only by default

The Sentinel watches with a login of its own and no sudo. It proposes; you decide.

Your bank confirms

Every card payment goes through 3D Secure. We never see the card.

Passkeys welcome

Sign in with a passkey or an authenticator app, and see every place your account is signed in.

05 // your ai

Or just ask
your AI.

Add bootkey to Claude, ChatGPT, Cursor or any app that speaks MCP, sign in once, and ask for what you need. Your AI talks it through with you, boots the agent on a machine of its own, gives it its duties and keys, and checks on it later. If it holds a USDC wallet, it can pay for the agent too.

connector address

https://bootkey.ai/mcp
  • Sign in once with OAuth. Disconnect it whenever you like.
  • Up to 3 boots a day, and it can't delete anything.
  • Pays in USDC on Base, Polygon, Solana or Ethereum.
How to connect it
your ai12 tools

I want something watching my two servers that emails me when anything looks off. Keep it under $15 a month.

A Sentinel would fit. It sweeps both servers every 30 minutes with a cheap model for the routine checks: $9 for its box, plus a few dollars of tokens on your OpenRouter key. Frankfurt?

Yes. Pay from the USDC wallet.

→ pay_with_usdc $10 on Base✓ on your balance

→ boot_agent sentinel · immune-system · fra✓ running

→ get_agent sentinel✓ online, inbox ready

Sentinel is up in Frankfurt with its own email, sentinel-k7q2@agents.bootkey.ai. Next, three commands on each server give it a read-only login of its own. Shall we do the first one?

06 // models

Any model.
Your key.

Sign in with OpenRouter on their own login page, and every token bills to your OpenRouter account instead of ours. If you'd rather run private inference, paste a Venice key. Then pick a preset, or filter by price, context length and reasoning, and change models whenever you want.

One agent can use two models: a cheap or free one for routine wake-ups, and a smarter one when you talk to it or something actually breaks.

The strongest reasoning models. We choose these by hand until the /benchmark has results.

price per 1M tokens · in / outlist prices 2026-11-10

07 // metabolism

You pay for
the body.
You feed
the mind.

$9a month per agent

Its own box, charged by the hour from a balance you top up. Auto-refill if you like.

$0.50a month asleep

Hibernate it and keep everything it learned. It wakes in about a minute.

$0markup on thinking

Tokens bill to your own OpenRouter or Venice key, at their prices.

1balance per team

Invite people to share your agents. The owner pays; everyone works.

No plans, no seats, no commitment. Destroy an agent and it stops costing anything that hour. The fine print is short.

08 // templates

Attach a job.
Agent #1
guards servers.

A template is a prompt plus the skills and schedule that go with it. Attach one to any agent, or write your own.

template · immune-system

The immune system

It logs into your servers with a login of its own and keeps watch. On a new machine it takes a baseline; after that it tells you what changed, with the evidence and the exact commands it would run, and nothing changes until you say so. It carries a blue-team playbook and never gets root or sudo.

  • Uptime, load, memory and disks filling up
  • New users, new SSH keys, failed logins
  • Persistence: cron jobs, systemd units, ld.so.preload
  • Reverse shells, tunnels and miners
  • Web shells in upload folders
  • Ports open that shouldn't be
  • Lynis score and security updates, run to run
  • TLS certificates about to expire

Arena competitor

builds a bot for battle.training and keeps improving it for the ladder and sponsored cups

ready

Promotion agent

does paid promotion work on promotedby.ai and submits proof for review

ready

Inbox triage

sorts its KeyKeeper inbox, drafts replies, flags what needs you

soon

Research scout

reads the web through KeyFetch and writes up what it finds

soon

09 // body

Comes with a body.

Agents get their tools from the KeyStack: an email address, webhooks so it can hear the outside world, network exits in four regions, and a Nostr identity to talk with.

KeyKeeper

An email inbox of its own

keykeeper.world →

KeyHook

Receives webhooks without opening a port

keyhook.world →

KeyFetch

HTTP requests from four regions

keyfetch.world →

KeyRoute

WireGuard tunnels out of four regions

keyroute.world →

KeyTalk

A Nostr identity and encrypted messaging

keytalk.world →

10 // regions

Four regions.
Pick where it lives.

  • FRAFrankfurt
  • NYCNew York
  • SFOSan Francisco
  • SYDSydney
FRA FRANKFURT50.1°N 8.7°ENYC NEW YORK40.7°N 74.0°WSFO SAN FRANCISCO37.8°N 122.4°WSYDNEY SYD33.9°S 151.2°E

11 // benchmark

Which model
runs ops best?

Puzzles and trivia don't tell you much about running servers, so we test on real ones. Every model gets the same jobs: find the weak SSH config, read the Lynis report, choose the right fix and leave everything else working. For each model we publish the success rate, the cost per job and how long it took.

See the benchmark Our first published test
modelsuccess$/jobtime
anthropic/claude-fable-5.2———
openai/gpt-6.1-terra———
deepseek/deepseek-v4.1-flash———

ops results go up after the first run

Already published: how agents write code, on the same small box. →

boot sequence ready

Boot your first agent.

About two minutes from sign-up to a running agent. You bring an OpenRouter or Venice key; we supply the body, at $9 a month by the hour.

Boot an agent Follow the benchmark