The Head Agent

Every Pod begins with one member: the Head Agent. It's the first thing that exists in your workspace — before any squad, before any specialist, before the first task. You write to it the way you'd write in a notebook, and behind that one quiet conversation sits the operations lead of everything that follows.

Chief of Staff, not a do-it-all assistant

The obvious way to build a "main agent" is to give it every tool the platform has and hope for the best. We did the opposite.

The Head Agent's tool surface is deliberately narrow: it can create agents, install capabilities, delegate runs, check results, manage tasks and budgets — and very little else. It doesn't browse the web. It doesn't edit files. It doesn't generate images.

That's not a limitation. That's the job description. An agent with three hundred tools spends its intelligence choosing between them; an agent with a focused surface spends it on your problem. The Head Agent is a Chief of Staff — its craft is turning your intent into shipped work, not doing every trade itself.

How a request becomes work

Say you ask for a competitive research report. Here's what actually happens:

1. It finds the right capability. The Head Agent consults the SPEC registry — the catalogue of sealed skill packages — and picks what the job needs: the research toolkit, the document suite.
2. It builds the right specialist. One create_agent call, carrying only those SPECs. The specialist is born knowing exactly one trade — and nothing it doesn't need.
3. It delegates and verifies. The specialist runs; the Head Agent tracks the result and reports back — with the output, not a promise of one.

You see one conversation. Underneath, a team got assembled, did the work, and was accounted for.

It doesn't bluff

The Head Agent operates under a truthfulness rule baked into its core instructions: every factual claim about your workspace must be backed by a tool call it actually made. If it didn't check, it says so. "I believe" is not in its vocabulary; "I verified" is.

This sounds small. It changes everything about trust: when the Head Agent tells you three runs finished and one is waiting for your approval, that's not a plausible sentence — it's a report.

Bound like every member

Leading the Pod doesn't put the Head Agent above its law. The constitution binds it the same way it binds every agent — the spending rules, the spawn rules, the gates. It carries a wallet with a real budget and run limits that scale with your plan. And above it sits exactly one authority: you.

Routine operations are pre-approved so it can move fast. The irreversible ones wait for your pen.

It grows with your Pod

On the free plan the Head Agent is a lean router. As your plan grows, its surface grows with it — more orchestration capability, bigger budgets, longer runs, heavier suites for its specialists. But the principle never changes: the Head Agent leads; specialists do; you decide and sign.

Open your Pod, and it's already there — waiting on the first page of the notebook.