A personal AI agent that runs 24/7 on its own server — with skills, persistent memory, a web dashboard, and Telegram access. Set up in minutes. No shared infrastructure.
Built by Julian Kaiser, the maker of Agents on Steroids.
You've seen the demos. An AI agent that reads your email, manages your calendar, runs code, searches the web. But getting there yourself means weeks of setup, broken cron jobs, scattered files, leaked API keys, and a system that forgets everything after every restart.
Great session, real progress — then the context resets and the agent asks "what are we working on?" All learning, gone.
API tokens in env vars, config files, agent processes. One prompt injection and your OpenRouter bill explodes.
Cron fires, session dies mid-task, nothing gets saved. You find out three days later when nothing happened.
You tell us what you need. We handle the rest.
We figure out your use case, set up your credentials, and configure the right skills for how you work.
A dedicated server, fully configured. Dashboard URL and Telegram bot ready to go — chat from your phone or browser.
The agent builds up knowledge about you and your work over time. Daily notes, memory consolidation, self-improvement — all automatic.
Search the web, read emails, check calendars, manage tasks, query databases, transcribe audio, generate speech — out of the box.
Persistent knowledge base that grows with every interaction. The agent knows your projects, your preferences, your context — and doesn't forget.
Web dashboard, Telegram, voice messages, file uploads. Your agent is always reachable — from your desk or your phone.
200+ models via OpenRouter, or bring your own Claude/ChatGPT subscription. Switch with one click. Your API costs, your account.
API keys are fully isolated from the agent. Even a prompt injection can't leak them. Encrypted tokens, hardened server, audit logging.
Dedicated machine, no shared infrastructure. You get full access. Cancel anytime — your data stays, no lock-in.
"You just give the agent these atoms — basic building blocks, small skills — and it figures out how to compose them into something useful. That's the real unlock."
Pay for Hostclaw management. Server costs go directly to Hetzner (~€4/mo). LLM costs go to your OpenRouter/Anthropic/OpenAI account.
Developers, founders, and power users who want a personal AI agent that actually works — not a demo. You should be comfortable with Telegram, have some LLM experience, and have a use case in mind. As Julian puts it: "It's a tool for tinkerers."
Any model on OpenRouter (200+), or bring your Anthropic Claude Pro/Max or OpenAI ChatGPT Plus/Pro subscription via OAuth. You switch models with one click in the dashboard. Your API costs are entirely separate from Hostclaw pricing.
Your agent runs on a dedicated server — no shared infrastructure, no multi-tenant anything. API keys are isolated from the agent process so they can't be leaked. The server is hardened and every API call is logged. You own the machine.
Depends on usage and model. With OpenRouter and a mid-tier model, most users spend €20–50/month. You can also use your existing Claude or ChatGPT subscription at no extra LLM cost. Hostclaw doesn't mark up API costs.
Yes. Full SSH access, your own machine. It can run on your Hetzner account or ours — your choice. No lock-in.
Cancel anytime. Your server and data stay — we just stop managing it. You can keep running the stack yourself. The Ansible playbook is all there is.
Weeks of setup, debugging, and infrastructure work. Hostclaw is the finished version of what you'd eventually build — battle-tested on daily usage. You skip the painful part and go straight to productive.
Book a call. If it's not a fit, I'll tell you fast.