Run your projects on servers you own.
A file in your repository says what runs. CI builds each commit; your server pulls the release, swaps it in with no downtime and turns the commit green, or red with the reason. No panel, no database: the file is the state.
- One Go binary
- Debian or Ubuntu, amd64 or arm64
- Open source, MIT
project = "crm" server = "nova-1" [apps.web] domain = "crm.example.com" release = "bin/rails db:migrate" worker = "bin/jobs" uses = ["main", "cache"] volumes = { uploads = "/rails/storage" } [apps.admin] path = "admin" domain = "admin.crm.example.com" uses = ["main"] [databases.main] [caches.cache] memory = "128mb"
# each value sealed with age to the server and to you recipients = [ "age1x7k9…2qpm", # nova-1 "age1m3dz…08fh", # you ] [apps.web] STRIPE_KEY = "age:YWdlLWVuY3J5cHRpb24u…" SENTRY_DSN = "age:LT4gWDI1NTE5IGpGdk1X…" [hakobu] status_token = "age:Ci0+IFgyNTUxOSBuUEFw…"
# written by your agent, from hakobu's init name: deploy on: push: branches: [main] jobs: deploy: permissions: contents: read packages: write # push to ghcr.io statuses: read # wait for hakobu/deploy uses: hakobudev/hakobu/.github/workflows/build.yml@v1.0.0 with: version: v1.0.0 image: ghcr.io/nova/crm
Nothing pushes to a server.
CI builds each commit into images and pushes them to a registry. The server checks the release tag every 30 seconds and deploys each new one. CI needs no access to it, and any CI that can push an image works.
- Your repositoryhakobu.toml, a Dockerfile per app, sealed secrets
- CI
hakobu build: each app, then the release image last - A registry
ghcr.io/you/app:main, or any other - Your serverpulls the release, deploys it, and sets the commit's
hakobu/deploystatus
The parts that guard your data, written once.
What's different in each project, its file, Dockerfile and migrations, is yours or your agent's to write. Deploys, backups, secrets and alerts are hakobu's, tested once for every project.
Deploys that can't lose data.
Migrations run in the new image while the old one serves. If they fail, nothing changed. A deploy never deletes: what the file drops is kept until a person deletes it.
- 1Planthe file against what runs; a typo stops here
- 2Dumpeach database the app uses, on the server
- 3Release
bin/rails db:migratein the new image - 4Healththe new version answers
/upnext to the old one - 5Switchtraffic moves over; no request is dropped
Back in one command.
rollback runs the release before, with today's secrets. Try it.
Secrets in the repository.
Each one sealed with age to the server and to you. Anyone, an agent too, can add one; only the two of you can read it. A diff shows which changed, never a value.
Backups that prove they restore.
Every database and volume, each day: sealed, copied to a second bucket with a key that may only write, then restored into scratch before it counts.
- crm/main: read back and restored, 21 tables
- crm/web/uploads: archive read through
- Your key opens them too, without the server.
One email when it breaks, one when it's over.
An app down, a failed deploy or backup, a full disk, or the hourly audit finding a port or container that shouldn't be there. A pulse to a dead man's switch says hakobu itself is alive.
Agents welcome. Data stays a person's.
Each server is its own MCP server, over SSH: an agent sees status and logs, deploys, rolls back and sets a repository up, with a skill for Claude Code. Nothing to install on your laptop. Restoring or deleting data is root's, and a person's at a terminal.
[caches.sessions] and gave web SESSIONS_REDIS_URL. c3d9a20 is live on nova-1.ssh -t root@nova-1 hakobu restore crm/main, at your terminal. My key can't.Nothing comes in.
The server pulls its releases, hakobu's own too. You and your agent reach hakobu over SSH, with keys that run its commands and nothing else; its API listens on loopback alone, and the web reaches your apps on 80 and 443 for their own domains, or through a Cloudflare Tunnel with no port open at all. The audit mails you about anything else that listens.
Yours, all of it.
hakobu calls no provider's API: no DNS, no cloud, no Git host beyond the commit status. You bring the pieces, and can swap any of them.
- A Linux serverDebian or Ubuntu, amd64 or arm64. The installer sets up rootless Docker.
- SSHYour laptop has it: you and your agent reach hakobu through it, and nothing else.
- A registry and CIghcr.io and GitHub Actions work with no setup.
- An S3 bucketR2, Hetzner, AWS or any other, for backups.
- An SMTP accountFor the alerts.
- Your owner's keyOpens every secret and backup, even without the server. The server makes it, for your password manager.
From a bare server to a green commit.
-
1
A server
One command; it asks the server's name and how visitors reach your apps (your agent gives them as variables). It ends with the line that gives your agent the server's MCP. After that it updates itself. A server
$ ssh -t root@203.0.113.7 'curl -fsSL https://hakobu.dev/install.sh | bash' $ claude mcp add --scope user nova-1 -- \ ssh -T deploy@203.0.113.7 mcp
-
2
A project
Ask your agent to set the repository up: it writes hakobu.toml, the workflow and its skill. You type the secrets in your own terminal, sealed on the server; the agent adds them, never seeing one. A project
you> Set this repository up for hakobu on nova-1. $ ssh -t nova-1 secret set crm hakobu status_token you> I sealed the status token; add it and push.
-
3
Push
CI builds the apps and the release; within 30 seconds the server deploys it, and the job turns green once it's live.
$ git push $ ssh nova-1 status nova-1 ● all good hakobu v1.5 · Debian 13 · arm64 crm deployed a1f3c9e · 1 min ago ● admin up ghcr.io/nova/crm-admin ● web up ghcr.io/nova/crm-web ◆ cache valkey 12 MB of 64 MB ◆ main postgres 1.2 GB · backed up 6 h ago disk ━━━━━━━━──────────── 41% mem ━━━━━━━━━━━━──────── 61%
I build hakobu on my own, and it runs my apps and a client's today. Every feature on this page is one I needed on a real server first.