Two service profiles, and how much memory each one uses on the machine.
- Default Setup 1151 MB essential services running.
- Full Setup 3,594 MB all services enabled
- Measured On 3,915 MB total memory of a 4 GB machine.
Real Linux labs for your students, under your brand. Give your AI agent a real computer. Your own Linux machine in the cloud, ready in your browser. Run your team's apps and databases on one server you control. The whole platform on your own server, behind your own sign-in. Give every department its own labs, with limits you set.
Every workspace is a real Linux machine with a public URL, a database beside it and an app store in front of it. Your people use it. Your AI agents drive it through one MCP connection.
How the trial works: sign in, tell us about your institution, and once we approve it you get a dedicated 4 vCPU / 8 GB server under your own brand, free for 30 days. Afterwards the same server carries on as the Nano plan at €30.00 a month, and you can move to a bigger machine whenever you want. Want us to set it up for your team? Done-for-you setup →
Your AI used to hand you code to copy and paste. Here, it runs it for you.
Chat that can use a computer
What if your AI had a computer?
Chat used to be: ask something, get an answer. Connect Claude, Codex or any MCP client to your platform and the same chat gets a real Linux machine. It installs, runs, fixes and ships, and you watch every step.
- Before
- It told you the commands. You ran them.
- Now
- It runs them, on a workspace with its own address, database and credentials.
- Always
- Every call is logged against your account, and your role decides what it may touch.
Put my notes app online with a Postgres database.
On it. Working on your platform:
deploy_labNode workspace, SSH readyrun_commandnpm install && npm run buildcreate_databasePostgres, its own credentialsconfigure_lab_routingPort 3000 on HTTPS, certificate issued
Live at https://notes.your-domain. Want a custom domain next?
Built for What You’re Building
Runs on our own bare metal in Hyderabad, India, and 4 more regions on request · your data stays on your workspace · no per-seat licensing
390+ MCP serversFrom the store: thousands of tools through one secure gateway, on one connection. Role, tier and module decide what a client is shown, and every call is audited.
114+ appsIn the store. Fork one, version it, build it with logs, publish it back.
6 enginesManaged databases and queues beside every workspace, with per-user credentials.
165sFrom the click to a platform serving your people, built from a ready image.
Powered by this platform
Already in production, at scale
-
selfmade.ninja
- 16K
- students
The Selfmade Ninja Academy.
-
labs.selfmade.ninja
- 22K+
- lab users
- 125+
- labs
The academy's own labs, on this platform in production.
-
Government of Tamil Nadu
- 2000+
- players
- 200+
- CTF machines
A capture-the-flag event we hosted on our labs, run together with the police.
One connection, many environments
Your agent gets a platform, not one throwaway box
An agent with one box writes code. An agent with a platform ships products. Give it a workspace per project, each with its own address, its own database and its own credentials, and let the platform do the plumbing.
Your agent
Claude, Codex or any MCP client, holding your permissions
deploy_labDeployA real Linux workspace from any template in the store, with SSH and a VPN address, in one call.
configure_lab_routingRouteA public URL and a certificate for any port. Custom domains too. No reverse proxy to write.
create_databaseConnect a databaseMySQL, PostgreSQL, MongoDB, Valkey or RabbitMQ, with credentials only that workspace holds.
publish_templatePublishTurn a working workspace into a template. Version it. Put it in the store for the next agent.
claude mcp add --transport http labs https://<your-domain>/mcpEvery call is logged against the account that made it. Quotas stop an agent at exactly what the plan allows. Assessments are off limits to agents by design.
Sized honestly
Run what you want. Nothing else.
Switch on only the modules and services your platform needs, and see how much memory each profile uses before you deploy.
Modules · what people see
2 always on / 6
-
Labs
always -
Instances
always -
Learn
optional -
Discuss
optional -
Evaluate
optional -
CTF Missions
optional
Service profile · default
6 lit / 17 bays
- traefik
- mq
- labsctl
- mysql
- postgresql
- memcached
- adminer
- judge0
- dvwa
- wgvpn
- labs
- mongo
- mariadb
- valkey
- rabbitmq
- redisinsight
- embed
The console
Deploy, snapshot, console, done
Start workspaces, take restore points, open a live serial console and watch the deploy log as it runs, in the console we run our own fleet from.
The app store
Browse. Install. Publish back.
114+ templates, from a plain Ubuntu box to Kali with a desktop, n8n, Jupyter, Qdrant, MinIO and the labs your own people build. Every one is a Dockerfile with a config, versioned in Git.
- Ubuntu, desktop
- Kali Linux, desktop
- Docker
- n8n
- Jupyter
- Qdrant
- MinIO
- Node-RED
- Zephyr RTOS
- Flutter
- Buildroot
- ESP-IDF
- Java desktop, desktop
- MySQL Workbench, desktop
- RedisInsight
- nginx RTMP
- ZealPHP
- U-Boot
- GCC 3
- Essentials
- and 94 more
A dot marks a template with a desktop in the browser.
Fork and version
Fork any template, change it, build it, read the build log. Every version is kept and any one can be restored.
Publish to your store
A template your team built becomes an app your organization installs. Publishers can be verified.
GUI when you need it
Templates can carry a full desktop in the browser. Kali, Ubuntu and Java desktops ship in the catalog.
Routing and TLS, handled
Every workspace gets an address the world can reach
A public URL and a certificate per workspace, on your domain or ours. Nobody writes a proxy config, and nobody renews a certificate.
| Host | Upstream | Certificate | Replicas | Health |
|---|---|---|---|---|
| notes.arun.selfmade.codes | web :3000 | wildcard, DNS-01 | 3, round robin | healthy |
| api.example.comcustom domain | api :8080 | issued, TLS-ALPN | 1 | healthy |
| pg.arun.selfmade.codesport 9443, SNI | tcp :5432 | wildcard, DNS-01 | 1 | healthy |
Domains and certificates
Platform subdomains and custom domains, both with automatic certificates. Up to 25 custom domains per user.
Ports, not just HTTP
Expose TCP ports with TLS and SNI, up to 10 per workspace. Databases and game servers reach the internet safely.
Replicas and rollback
Run replicas behind one weighted round-robin service, with health checks. Every deploy is recorded and any one can be rolled back.
What you get
A platform, not a plan page
We handle routing, TLS, VPN, credentials and limits. Run it on our bare metal or yours.
Workspace
The platform
Governance
Real Linux, yours
A container with root, SSH, a WireGuard address and VS Code in the browser. Persistent, not a session.
The app store
114+ templates your people fork, build, version and publish. The reason the platform exists.
Organizations and departments
Departments under an organization, each with its own members and its own ceilings.
Public URL and TLS
A certificate per workspace, custom domains, TCP ports with SNI, replicas and rollback.
Managed services
Six engines with per-user credentials: MySQL, MariaDB, PostgreSQL, MongoDB, RabbitMQ, Valkey (Redis-compatible).
Limits you set
Storage, workspaces and lab quotas per department, enforced rather than reported.
Console and serial access
A real serial console in the browser, so a machine you have locked yourself out of is still reachable.
One MCP connection
The agents your people already use, with the permissions those people hold. Thousands of tools through one gateway, from 390+ MCP servers in the store. Each client sees only what its role allows. Every call is audited.
Your identity provider
You choose it and the OAuth application is yours. Nobody signs in through us.
Where your workloads run
One region serving today, four more on request
A dedicated server takes 24 to 48 hours to stand up, so a new region is a conversation and a lead time, not a checkbox. We would rather say so than claim one we do not have.
Hyderabad, India
Serving today
UAE
On request
USA
On request
Kenya
On request
Oman
On request
The control plane is separate
The control plane runs independently from the compute fleet, so an issue with one node affects only the workspaces running on that node.
Bring your own hardware
Already have a server or datacenter? The same platform can run on your own infrastructure.
Dedicated network addresses
Every workspace can receive a dedicated public IP from our routed subnet, with 1:1 addressing when required.
Resources
Learn it, build on it
The documentation and the running academy, all readable before you sign up for anything.
MCP & the API
Connect Claude, Codex or any MCP client to thousands of tools: 390+ MCP servers from the store, through one secure gateway. It sees only what your role is allowed to call.
Platform guides
Deploying labs, building templates, networks, storage and the console.
See it running
The platform you would be buying, in production, with real courses on it.
GitLab
Lab templates and the tooling around them. Needs a GitLab sign-in.
Talk to us
An organization that outgrows the plans gets a whole dedicated server.
Partners
Resell it, or run it for someone else
Agencies and consultancies who set up platforms for other organizations: you hold the relationship, we hold the hardware. Talk to us about terms — there is no self-serve partner programme yet, and we would rather say so.