OPOS is an AI-native operating system for your company — a plain GitHub repo where your mission, your departments, your roles, and your work all live as readable files, run by AI agents.
In one paragraph: OPOS is an AI-native company operating system distributed as a free Copier template — a GitHub repository of markdown and JSON files that encodes a company's mission, policies, departments, agent roles, and work-in-flight backlogs. It runs on Claude Code's native primitives — agents, skills, and folder conventions — with no runtime and no orchestration code. One founder, a git repo, AI agents doing the work, and human oversight through pull requests: every change lands as a reviewable git commit before it counts.
Why now: AI agents have made the org chart software. The work a department used to do can now be described, delegated, and executed by an agent. OPOS is the bet that the next great company is one founder, a git repo, and an organization that runs itself.
A company is really just a set of decisions, roles, and repeatable work. OPOS encodes all of it as markdown in a single repository: who you are, what you're building, how the work gets done. There's no platform to log into and no runtime to operate — your company is the repo. You lead in plain language; an AI organization turns your intent into execution.
Because it's all files in git, your company is auditable, forkable, versioned, and entirely yours. Nothing is locked inside someone else's SaaS.
For most of history, ambition was capped by headcount. To do more, you hired more — and hiring is slow, expensive, and hard to undo. OPOS removes that ceiling. When the org chart is software, a single person can hold the leverage of an entire company: a strategy function, a finance function, an engineering function, all staffed by agents that draft, analyze, and execute while you decide what matters.
This is what the "one-person company" actually looks like in practice — not a founder doing everything, but a founder directing everything.
You open your repo and talk to your company in ordinary language:
You: Let's ship the new pricing page this week.
OPOS: Spinning that up — opening a tracked task, drafting the plan, pulling in engineering. I'll check in before anything ships.
You: Should we expand into the EU next quarter?
OPOS: That's a strategic call — let me run it past the relevant leads, weigh the tradeoffs, and bring you a recommendation to approve or reject.
You state the goal. The organization decomposes it, does the work, and comes back when it needs your judgment.
- Your company is markdown, not a platform. No runtime, no lock-in. Open the repo, read everything, own everything.
- An organization that grows itself. The org chart isn't hardcoded. As needs emerge, OPOS can design new roles, new departments, and new processes — so it keeps pace with your company instead of going stale.
- AI-first by principle. New capabilities default to AI. Humans are brought in deliberately — where lived experience, legal accountability, or physical presence genuinely require a person.
- Autonomy with a human in the loop. Routine work can run on its own schedule, but the decisions that count — shipping, spending, going public, hiring — always come back to you.
- Everything in git. Every decision, every change, every piece of work is tracked, reviewable, and reversible.
This isn't an agent framework for engineers to wire together pipelines. It's an operating system for running a company — built for founders, not just builders.
Unlike Python orchestration frameworks — CrewAI (57.7k GitHub stars as of August 2026), MetaGPT (70.1k stars), and Microsoft's AutoGen (60.7k stars, now in maintenance mode) — OPOS ships zero orchestration code. The company itself is the artifact: readable markdown files, versioned in git, executed by Claude Code's native agent and skill primitives. And unlike hosted no-code agent platforms such as Lindy (1,000+ integrations, plans from $29.99 per user per month) or Claude Code swarm harnesses like claude-flow, now Ruflo (69.5k stars), OPOS is a free, forkable template where every agent, policy, and decision lives as an auditable file.
As LangChain's June 2026 framework review puts it, "The best agent frameworks give developers clear primitives for tool calling, state management, and inter-agent communication without hiding what's happening underneath." OPOS applies that principle organizationally: Claude Code supplies the primitives, and OPOS supplies the company-shaped conventions layered on top.
OPOS is early and moving fast. Today it gives a single founder a credible, self-organizing company in a repo; the trajectory is a platform that covers ever more of what a real organization does — more of the company describable in files, more of the work the organization can carry on its own. The surface area keeps growing. This README stays deliberately high-level because the capabilities won't stop expanding.
OPOS is distributed as a template you scaffold in one command, then bring to life in conversation:
copier copy gh:Koroqe/OPOS my-company-os -d COMPANY_NAME="My Company"Open the new repo in your editor, start a Claude Code session, and talk to your company — it'll walk you through setting up your mission, departments, and first goals.
In OPOS, the entire organization — mission, policies, departments, roles, and backlogs — exists as markdown files in a GitHub repository, executed by Claude Code agents under human oversight. The founder directs work through a chief-of-staff agent; departments such as R&D, Finance, Legal, and Commercial each have a lead agent with a charter. The framing: not a founder doing everything, but a founder directing everything.
CrewAI (57.7k GitHub stars), MetaGPT (70.1k stars), and AutoGen (60.7k stars, now in maintenance mode) are Python frameworks: you write orchestration code. OPOS ships zero code — departments, roles, and processes are markdown files executed directly by Claude Code, so a non-engineer can read and change the whole company definition. Full breakdown: OPOS vs CrewAI vs AutoGen vs Lindy.
Lindy is a hosted no-code platform with 1,000+ integrations from $29.99 per user per month — convenient, but your automations live in its cloud. claude-flow, now Ruflo (69.5k stars), is a swarm meta-harness for Claude Code. OPOS is neither: a free, forkable Copier template where the company itself — org chart, policies, decisions — is a version-controlled git repository you own.
Install Python 3.10+, Copier 9.0.0+ (pip install copier), git, and the authenticated gh CLI, then run the copier copy command from Get started above, followed by git init, an initial commit, and gh repo create --private --push. CORE files (agents, skills, templates) arrive pre-installed; STARTER files are yours to customize and are never overwritten by copier update.
Run the /company-setup skill in a Claude Code session at the repo root. It fills the six content-blank areas — Mission, Values, strategic priorities, the six default department missions (rnd, finance, people, legal, commercial, pr), and policies — in about ten questions and fifteen minutes, writing progress to files as you go. Then browse the populated OS via /serve-console at http://127.0.0.1:8765/.
Yes — parallel sessions are first-class. Each terminal's chief-of-staff steward registers its own GitHub issue via task-register, the .current-task file lists every active issue number, and task-update/task-complete auto-pick with one active task or require --issue with several. task-pause and task-resume work per-issue.
The chief-of-staff steward follows a graduated permission ladder — auto (reads, tests, drafts), notice, confirm (e.g. git commit, closing issues, creating agents), explicit approval (e.g. git push, release cuts), and hard refuse for destructive operations. Shipping, spending, and strategic pivots always come back to you. The full ladder is documented in the chief-of-staff agent charter inside the repo. More questions: the full OPOS FAQ.
- Founder's guide — the full, hands-on walkthrough of running your OPOS company ships inside every scaffold as its
README.md(source:README.md.jinja). - Known limitations & roadmap —
RISKS.mdlays out, honestly, what's solid today and what's still ahead. - OPOS vs CrewAI vs AutoGen vs Lindy — where a git-based company OS sits among agent frameworks and no-code AI employees.
- OPOS FAQ — running a one-person company with AI agents, departments as markdown, and oversight via pull requests.
- How to run a company as one person with AI agents — six approaches compared, from no-code AI employees to company-as-code.
Built on Claude Code. Early-stage and evolving — feedback and forks welcome.