Real root, real VM
Install Docker, run systemd, build a kernel. It's a Linux machine (QEMU/KVM), not a container with guardrails.
Real Linux VMs with full root — for you and your AI agents.
An open-source, self-hosted alternative to Replit and Codespaces that runs on your hardware — and, through its MCP server, an agent sandbox like E2B. A browser IDE and a built-in agent for you; ten task-shaped tools for the agents you already use.
Three commands. Your hardware. No per-seat billing.
$ git clone https://github.com/HectorPulido/pequeroku.git $ cd pequeroku/source && ./start.sh ✔ lab online → http://localhost/dashboard/
A real Debian VM, opened from the browser. Phone, tablet or laptop — no setup, no waiting.
Install Docker, run systemd, build a kernel. It's a Linux machine (QEMU/KVM), not a container with guardrails.
No cold starts: your workspace is ready when you are, and resets to a clean slate when you want one.
Pequenin edits files, runs your tests, starts servers and verifies they respond — live, inside the VM. Bring your own OpenAI-compatible provider.
Claude Code, Cursor, or your own bot get disposable root VMs over MCP. The blast radius is one VM — not your laptop.
The IDE is responsive — laptop, tablet, phone — with a live preview of whatever your VM serves.
Open source, hackable at every layer, cheap to run on a homelab — with quotas and roles for when you share it with a team.
Everything the dashboard does is a versioned public API — and the MCP server is just
a client of it. Point any MCP-capable agent at /mcp and it
can create VMs, run code, move files and inspect ports on its own.
API keys carry scopes (read < exec <
admin), destroying needs confirmation, and runs carry a
timeout and TTL. Sane defaults, one isolated VM of blast radius.
10 task-shaped tools, plus ready-made prompts and on-connect instructions.
claude mcp add --transport http \
pequeroku http://localhost/mcp \
--header "Authorization: Bearer pk_xxx"Works with Claude Code, Claude Desktop, Cursor, opencode — anything that speaks MCP.
curl -X POST http://localhost/api/v1/runs \
-H "Authorization: Bearer pk_xxx" \
-d '{"command":"python main.py",
"files":[{"path":"main.py",
"content":"print(\"hi\")"}]}'
# → {"status":"succeeded","stdout":"hi\n"}One-shot: boots a throwaway VM, runs, returns output, destroys it. OpenAPI spec at /api/v1/schema/.
from pequeroku import PequeRoku
pq = PequeRoku(api_key="pk_xxx",
base_url="http://localhost")
print(pq.run("echo hello").stdout)A thin Python SDK over the same API — no privileged side paths.
git clone https://github.com/\
HectorPulido/pequeroku.gitcd pequeroku/source
./start.shIdempotent — generates secrets, enables KVM if available, runs Docker Compose.
open http://localhost/dashboard/Log in, create your first VM, or let the agent scaffold the project.
Needs Docker Compose. A Linux box with /dev/kvm gives fast VMs — a homelab is plenty.
| PequeRoku | Cloud IDEs | Agent sandboxes | |
|---|---|---|---|
| built for | humans + agents | humans | agents |
| hosting | your hardware | their cloud | their cloud |
| compute | real VMs (KVM) | containers | microVMs |
| root | full | limited | ephemeral |
| lifespan | always-on | boot & wait | minutes |
| cost | your electricity | per seat | per second |
Cloud IDEs: Replit, Codespaces, Gitpod · Agent sandboxes: E2B, Daytona
Free forever. GitHub issues, the wiki, and a maintainer who replies to everything.
Want PequeRoku deployed on your infra, tuned for your team, or extended with custom features? I do that.
“I built PequeRoku because I couldn't find a dev platform that was open, root-access, always-on and affordable to run myself. If you want it running in your rack — I'll help you get there.” — Héctor Pulido, author
$ ./start.sh