SELF-HOSTED · SWITZERLAND · NO TRACKING
92 AI models and a real Linux server, in one account. The AI installs software, writes code, runs services and puts them online for you — with root, in your own container, on Swiss hardware. You approve each command and see what it costs as it happens.
APPROVAL REQUIRED
$ npm install --save-dev vitest
That is the product: the command before it runs, the price as it is made.
01
Claude, GPT, Gemini, DeepSeek, Kimi and 87 more — one login, one balance, no separate contracts to sign. Switch model mid-conversation and keep the thread. Tokens are billed at the provider's own price with no markup; topping up adds a 10% service fee and a 3% currency buffer, and every answer shows its cost to the centime while it streams.
02
Root in a real Ubuntu, Debian or Alpine container — apt install,
compile, run a database, start a web server. Publish it on *.zerolog.ch in one
click, or point your own domain at it. Files, ports and terminal are yours to open at any
time.
03
Type /goal and the agent keeps working for hours — installing,
testing, fixing, retrying — while you do something else. Up to 12 hours a run on Pro, unlimited
on Max. Every step is logged, every command waits for approval, and the terminal is yours to
take over the second you want it.
Every shell command the AI wants to run appears first, in full. Allow it, edit it, or refuse. Nothing touches your container without you saying so — and you can put trusted commands on an allow-list once you are tired of clicking.
Input tokens, output tokens and the price appear while the reply streams. Set a hard spending cap per chat or per account; when it is reached, work stops instead of quietly continuing.
Upload into the container, browse what the AI wrote, download the result. A file tree and an editor, not a chat window pretending to be a filesystem.
Open a shell into the same container the AI is working in, mid-task. Fix the one thing yourself and hand it back — the conversation keeps the context.
Expose a port and get an HTTPS URL on *.zerolog.ch
immediately. On Pro and above, point your own domain at it.
Give an agent a job on a schedule — a nightly build, a daily scrape, a weekly report. It runs while you sleep and the log is there in the morning.
Anonymous chats are encrypted in your browser: the server stores ciphertext it cannot open. The prompt still reaches the model to be answered — encryption protects what is stored here.
Have a provider account already? Use your own key and pay them directly. Keys are encrypted at rest and never shown to the browser again.
Pick a task. The AI proposes a command, you allow or deny it, and the cost counts up as the answer streams. This is a scripted walkthrough of the real interface — no account needed.
Pick any model. The chat carries your container, your files and your credentials.
Before a command runs you see it, and you can allow it, edit it, or refuse.
Tokens, cached tokens and the price land in the ledger while the answer is still streaming.
You can build this on a hyperscaler. It takes an afternoon of IAM roles, VPCs and security groups, a separate contract with each model provider, and a bill that only tells you the truth at the end of the month.
Sign up, type a sentence, and the AI is working in a container. No roles, no policies, no networking to design first.
Credit is prepaid and spend limits are hard. When it is gone, work stops — it does not keep running and invoice you in four weeks.
Each answer shows its own cost while it streams. Not a line item in a console you have to go looking for.
A comparable on-demand instance at a hyperscaler costs several times that before you have added a single model call.
Elsewhere the model service and the compute are two products you wire together. Here the model already has root on your box.
Hosted on our own machines in Switzerland. No tracking, and your chats are never used to train anything.
ONE ACCOUNT, EVERY MODEL
Switch between them mid-conversation. The thread comes with you.
/goal port this Express app to Fastify
The agent reads the routes, rewrites them, runs the test suite, fixes what it broke, and leaves a diff to review in the morning.
/goal scrape this data daily and mail me a summary
A scheduled task, a small database in the container, and a public URL for the dashboard. No servers to rent, no cron on your laptop.
Set up Postgres and seed it from this CSV
Root in a real container means the AI installs the database instead of explaining how you could install the database.
Put this Next.js build online on my domain
Publish a port, point your domain, TLS handled. From folder to live URL without touching a cloud console.
The short version of the manual: what to type, where to click, and what happens next.
Start a chat and pick a model from the selector above the message box. You can change it mid-conversation and the thread comes with you, so you can start cheap, switch to a stronger model for the hard part, and switch back.
Open Container, click Neu, choose a base image (Ubuntu 24.04, Debian 12 or Alpine) and a name. It boots in seconds and stays yours until you delete it, so files survive restarts.
apt install, build from source, run a
database. It is a normal Linux box in a sandbox.When the AI wants to run something, the command appears in full and waits. Three choices: Erlauben runs it, Bearbeiten lets you change it first, Ablehnen refuses and tells the model why.
ls and cat? Put safe
commands on the allow-list and only the interesting ones stop.Type /goal and what you want, in plain language:
/goal migrate this project to TypeScript and get the tests green. The agent
plans, works, checks its own result and keeps going until it is done or it hits something
it cannot solve.
The Dateien tab is a real file tree of the container. Browse what the AI wrote, open it in the editor, upload your own, download the result.
Start something listening in the container, then publish the port from
the Ports panel. You get an HTTPS address on *.zerolog.ch straight
away, with the certificate handled for you.
In Aufgaben, give an agent a job and a time: a nightly build, a daily check, a weekly summary. It runs on its own and leaves a log.
Every answer shows its own cost while it streams: input tokens, cached tokens, output tokens, price. The ledger under Verbrauch keeps the full history per chat and per model.
Switch a chat to anonym and it is encrypted in your browser before it is sent. The server keeps ciphertext it cannot open, plus the billing counters.
Already paying a provider directly? Add your key under Einstellungen and that usage goes to your account there instead of your ZeroLog credit.
Ask for an image in a normal message. The AI picks a suitable model or you name one: fast and cheap for drafts, the strong one for detail, and one that is good at text inside the image.
Install the app, open Geräte on the website and tap Gerät koppeln. Scan the code with the app and it is signed in.
A plan buys hardware: more cores, more memory, more containers running at once, longer agent runs and public ports with your own domain. Model usage comes out of prepaid credit at the provider's own token price. You upgrade when you want a bigger machine, not because a meter ran.
Prices in CHF. Nothing is charged automatically — credit is topped up when you choose to.