Memory
A structured knowledge base
Strategy, offers, personas, pipeline, and decisions live as human-readable markdown in one place — not scattered across ten tools and your memory. The source of truth is readable without an agent present.
Loop OS / the operating system behind Hooman Loop
Loop OS is an AI operating system for the business — a self-hosted agent layer on a structured knowledge base, running on a Mac mini. Data stays local, the system learns, and you start with one project, not the whole stack.
The problem
Strategy lives in a deck, the pipeline in a spreadsheet, context in email, and half the decisions only in your head. Chats disappear, tools do not talk, and every question starts from scratch. Loop OS is the intentional layer on top: one place that holds the business and informs everything downstream.
How it is built
Memory, agents, and surfaces — a practical stack, not a magic box. For the team-safe version of this model, read the agentic operating model.
Memory
Strategy, offers, personas, pipeline, and decisions live as human-readable markdown in one place — not scattered across ten tools and your memory. The source of truth is readable without an agent present.
Agents
A self-hosted agent layer runs tools, research, file work, and scheduled routines locally. Named roles do the labor; you keep taste, approval, and commitments.
Surfaces
Capture fast in Slack or Telegram; work lands in briefs, tickets, reports, and operating views you can actually run the business from.
Named roles
Loop OS is a real operating layer with named roles — not a single assistant. Each has a job and an approval gate.
Coordinates — routing, triage, delegation, and follow-through.
Research and relationship memory — reports, entities, and synthesis.
Design, strategy, and domain drafting when the work needs depth.
Taste, approval, and final commitments stay human.
In real work
01 / Bug-fix loop
A screenshot in Slack turns into a tracked issue, an implemented fix with a demo attached to the PR, and a ping back to Luis for review — without leaving the operating layer.
02 / One repo, one truth
The business model, personas, competitor swipe, testimonials, and offers live in one repo. When the strategy changes, the content, offers, and outreach downstream stay aligned.
The stack, high-level
An always-on local machine, an agent runtime, durable memory, chat surfaces, and the tools a working practice already uses. Modular — start with one project and a need.
Self-hosted agent orchestrator on a Mac mini. Data stays local and portable.
Fast capture and status — the chat surface into the system.
Code access and task execution for the engineering loop — Cursor CLI, Claude Code, or OpenAI Codex.
Issues and project state the agents read and write.
Shared tokens and components the agents build against.
The durable, human-readable source of truth underneath it all.
The Hooman Loop
Loop OS started as Hooman Loop's own operating layer. The same thinking — knowledge base first, agents second, human judgment always — is how we help teams build theirs.