HQ is a file-based AI operating system that runs three businesses at once. A single founder delegates to eight specialist agents, who act inside real tools, remember every decision between sessions, and stay inside hard guardrails.
A single founder runs a software house, a global membership, and an AI executive program. The system gives them a C-suite that never sleeps — and never forgets.
Work flows down from a human intent into a team of agents, through real-world tools, against a persistent memory — all wrapped in guardrails.
You. One human sets intent and direction — "what," not "how."
Eight specialist agents that delegate, draft, sell, build and coordinate.
Each agent works within the right brand voice, persona and data.
Agents act in the real world through Model Context Protocol connectors.
Markdown files hold decisions, goals, plans and wisdom — memory that compounds.
Two AI engines run the same repo: Claude Code and OpenAI Codex.
Cloud-safety gates, credential protection, write-hooks and audit logs.
The CEO delegates; the rest execute. Clear borders mean no two agents step on each other's work.
Strategy, delegation and the hard calls. Breaks complex work across the team and synthesizes the answer.
Execution, processes and automation. Turns chaos and "no time" into a plan that ships.
Sales, pipeline and conversions. Lives in the CRM, chasing deals that go quiet.
Marketing, content and lead generation — always written in the founder's voice.
Integrations, automations and technical security. Builds the rails the team runs on.
Replies as you — emails, DMs, comments — in your exact style. Indistinguishable.
Runs the local mailbox end to end: scan, draft, send and archive by a strict protocol.
Maintains the repo, triages the roadmap and routes work to the right thread.
The system detects which world it's in — and switches voice, language and data accordingly.
A software house: clients, delivery, pipeline and a consolidated P&L. The financial engine room.
A global community: Trust Circles, Navigators, Explorer and Lighthouse events. Always in English.
Cohorts, VIP masterclasses, GAIA and startcreating.ai — helping leaders build with AI.
Through Model Context Protocol, agents reach into the same tools a real team uses — with real data, never guesses.
The system's memory lives in plain files — readable, versioned, and shared between both runtimes. Nothing is lost when a session ends.
Every decision, goal and lesson is written down — so the system gets smarter every week, not just every chat.
Active, binding team decisions — checked before any non-trivial work.
A single source of truth for work in progress, capped at five.
Quarter, year and dream goals — the spine every task is aligned to.
A thought-inbox that triages raw captures: kill, delegate, automate — then do.
Insights from books, talks and podcasts — distilled into how the system behaves.
Safety comes from rules and limits — not from approving every single step. That's what lets the system move fast without breaking things.
Any work touching cloud, cost, secrets or deploys must pass a mandatory safety read first — least privilege, tagged resources, no surprises on the bill.
Secrets never land in the repo, prompts or logs — only the names of variables. No rotating or resetting keys without explicit consent.
Pre-tool hooks gate sensitive writes — sending mail, mutating the CRM — so a mistake is caught before it leaves the building.
Every tool call, correction and external action is logged — a full trail of what the system did and why.
Every change is measured against one question: does it move us up a level? Today the system sits at Level 6 — Self-Improving. Tap a rung.
The same repository is driven by two AI runtimes in parallel — pick the right engine for the task.
1M-context, adaptive thinking. Reads CLAUDE.md.
Parallel business runtime. Reads AGENTS.md.
Act and report back, instead of asking permission at every turn.
Safety through rules and limits, not step-by-step sign-off.
Track real metrics. Pull numbers from tools, never from memory.
Prefer CLI, APIs and Markdown — anything an agent can use.— Karpathy
Simple on the surface, ten levels deep underneath.— Parrish
Tell agents what to do; let them find the how.— Knight
One person commands eight specialist agents — the output of a team, the overhead of none.
File-based memory means decisions, goals and context compound between every session.
Agents work inside CRM, calendar, email and payments — not just in a chat window.
Guardrails, credential protection and audit logs make autonomy something you can actually deploy.
One human sets direction. The system does the rest — and keeps getting better at it.