The in-house AI toolchain, explained: what we set up and why each piece is there.

The quiet fear behind building AI in-house is usually this: “we cannot run our own AI infrastructure.” People picture servers, a platform to babysit, and a bill that grows while they sleep. None of that is what we set up.

The in-house stack is a handful of accounts your team signs into, the same way they sign into email. Nothing to host, nothing to patch. Here is the actual map, in plain language, so you know what you are committing to before you commit.

The fear it kills

“Infrastructure” sounds like a server room. In practice, an in-house AI stack is five categories of tool, each one a subscription you log into. The hosting, the updates, and the heavy compute all sit with the providers. Your team learns to point these tools at real problems. That is the whole job, and a curious operator can hold it.

The stack, in plain language

Five categories cover almost everything a team needs. We set them up together so you know what each one is for.

• A chat layer. The everyday assistant you already know. Ask a question, draft a document, think something through. It is where most work starts before it becomes anything more formal.
• An app builder. This turns an idea into a small internal tool with a real interface, the kind a teammate can use without knowing how it was made. This is where a workflow becomes an app.
• An automation layer. This connects your tools so a task runs on its own: a trigger happens, steps fire, an output lands. It is for the repetitive jobs nobody should be doing by hand.
• Cowork. A shared space where your team and AI work on the same task together, so the back-and-forth has a home rather than living in scattered chats.
• Managed agents. For the heavier jobs, an agent that can run a longer task with less hand-holding. You reach for this when a one-shot answer is not enough and you want something that works through a problem.

What to reach for when

The categories sound abstract until you map them to how work actually shows up. The rule is simple, and it scales with how often the thing repeats.

• A one-off question becomes a chat. You need an answer or a draft once. Open the chat layer and move on.
• A repeated task becomes an automation. If you are doing the same steps every week, that is a signal to hand it to the automation layer so it runs without you.
• A whole workflow becomes an app. When several people touch the same process and it needs a proper interface, you build it in the app builder. The managed agents and Cowork sit behind these for the heavier or more collaborative work.

Most teams live in the first two for months before they need the rest. You do not have to master all five on day one.

Why it lives in your team's names

Every account is set up under your own people, not ours. The subscriptions are yours, the apps your team builds are yours, and you own the IP. That matters for one plain reason: the capability stays with you. If we walked away tomorrow, nothing switches off and nothing is locked behind us.

With Setup, the year of subscriptions for both trained seats is included, so the running cost is visible and bounded from the start. After that the accounts are simply yours to keep.

The Run contrast

If you would rather not own any of this, Run is the honest alternative. The same kind of stack lives on our cloud, and your team uses it through a secure link. No accounts to manage, no tools to learn, no toolchain in your names. It is a A$1,500 discovery (credited) plus A$150 a month for hosting per workflow.

That is a fair trade when you genuinely want the result and not the wheel. The difference is only who holds the keys. With Setup the stack and the skill live in your building; with Run they live in ours, and you reach them through a link. Both ship working software. Pick the one that matches how much you want to own.