Any open-weight model you can host. The factory calls whatever endpoint you give it.

Yes — it runs entirely on your own machines. That is the point.

Where does our data go?

Nowhere. It never leaves your network.

Not ready to self-host?

Use shipd — the same system on our cloud. Get a token and deploy in a minute.

self-hosted · your infra · your models

Install the factory.

The factory is the agent system behind our products — the same one that runs shipd. Install it on your own box, point it at your own model, and it writes software from a plain-language brief, builds it, and ships it inside your walls: your model, your data, standard containers, no lock-in. No calls, no pilots. Install it and go.

install it →or use shipd on our cloud

01 / what it is

The system, in your walls.

→describe an app in plain language — the factory writes it on your own model, builds it with buildpacks, and ships it
→or let a meeting become the brief — point a bot at your call and the transcript ships as a working app
→or bring your own coding agent and just deploy the files — every path, one runtime
→each app runs as an isolated (gVisor) container behind your own gateway, with TLS
→apps that store data get a PostgreSQL database provisioned and wired in automatically
→standard OCI containers — export and run them anywhere, no lock-in
→your model, your data, your machines — nothing leaves your network

02 / install

One command, on your own box.

On a fresh Ubuntu server you control, as root, run:

install the factory

curl -fsSL https://aikaara.com/factory/install.sh | sh

No domain needed — it serves over magic IP-based wildcard DNS with real TLS. Prefer your own domain? Pass it, with *.apps.yourdomain.com pointed at the box:… | sh -s -- apps.yourdomain.com

It stands up the build-and-run system — Docker, gVisor isolation, Cloud Native Buildpacks, a TLS gateway — and a secure MCP server, then prints a claude mcp add line. The source is yours to read and keep.

Point it at your own model to turn on build — describe an app and the factory writes it on your model, then builds and ships it:

bring your model

export LLM_MODEL=openai/your-model
export LLM_API_BASE=http://host.docker.internal:11434/v1   # your endpoint
curl -fsSL https://aikaara.com/factory/install.sh | sh

Any OpenAI-compatible endpoint — a local vLLM or Ollama on the box, or your own hosted model. No model set? It still installs; deploy works and buildturns on the moment you add one.

03 / straight answers

Your models will be slower than the frontier.

True. On-prem open-weight models do not match hosted frontier models on raw capability. You run what you can host; the factory does the build-and-ship around it. Most of the capability, all of the sovereignty.

Why not just buy coding-agent licenses for our team?

Tools make your existing developers faster, and your data still leaves. The factory ships running software inside your walls — including where you do not have the developers.

What does it run on?

A Linux box with Docker. The installer sets up the build-and-run system; you point your coding agent at it. Size it to your workloads — start small, scale the host.

Who runs it in year two?

You do — it is standard containers on your own infrastructure. No vendor in the critical path, source you can read, nothing to renew to keep it running.

04 / faq

Which models?: Any open-weight model you can host. The factory calls whatever endpoint you give it.
Air-gapped?: Yes — it runs entirely on your own machines. That is the point.
Where does our data go?: Nowhere. It never leaves your network.
Not ready to self-host?: Use shipd — the same system on our cloud. Get a token and deploy in a minute.

install the factory or start with shipd on our cloud