Operator OS.
The context layer every operator needs before AI becomes useful.
By Yannic Desch · October 2026 · 5 min read
Stop repeating yourself to your AI.
Operator OS is a structured context kit for operators. Five markdown files. Any AI tool. Open source and runs on your machine.
Fork on GitHub
github.com/yannicdesch/operator-os
Every time you start a new AI conversation, you're starting from zero. You explain your role, your company, your priorities, your voice. You do it again the next day. Operator OS fixes that. Your AI reads the same five files at the start of every session — and arrives already knowing your business.
GET THE KIT + WHAT'S NEXT.
One email when a new skill or template ships for Operator OS. Plus the biweekly Builder First letter. No noise.
Why Operator OS?
Most AI prompts are disposable. You type, the AI responds, the session ends. Tomorrow, it knows nothing. Operator OS makes context persistent, structured, and reusable — across sessions, across tools, and eventually across your whole team.
Stop re-briefing your AI every session.
A VP who re-explains their role, their company, and their priorities every conversation loses 30 minutes a week. That's 26 hours a year. Operator OS is the briefing file your AI reads before you say a word.
Better output from any model.
Context is the variable that matters most. The difference between a generic AI response and a response that matches your voice, your numbers, and your business situation isn't the model — it's the context you give it. Operator OS makes that context explicit, structured, and always available.
One context file. Every AI tool.
Claude, ChatGPT, Gemini, Cursor, Perplexity — Operator OS works in all of them. Add it as a Project file, attach it to a system prompt, or paste it at the top of any session. Same context everywhere.
Team context that doesn't live in someone's head.
When a new EA, consultant, or agent joins your workflow, they read Operator OS. They know your company, your priorities, your communication style. Onboarding time: zero. The institutional knowledge that usually takes weeks to transfer moves in minutes.
Who it's for
Operator OS is for operators who use AI daily and are tired of the context overhead.
You use AI to prepare for board reviews, draft roadmaps, write strategy memos, and think through product decisions. Operator OS gives your AI your product context, your decision log, and your communication style — so every output is already calibrated to your situation.
You use AI to synthesize reports, prep for investor calls, draft comms, and think through strategic options. Operator OS gives your AI the company numbers, the open decisions, and the key stakeholders — so it can respond like an EA who's been with you for two years.
You use AI to process information faster, draft faster, and think more clearly under time pressure. Operator OS gives it your process context, your team structure, and your current priorities — so it acts on what actually matters to you, not a generic interpretation of your request.
How it works
Five files. Each one answers a question your AI otherwise has to guess. Together, they give any AI model the same situational awareness a good CoS builds over months.
The operating principle
Operator OS doesn't change how your AI works. It changes what your AI knows. A capable model with the right context consistently outperforms a better model without it. The bottleneck was never the model.
The files
Each file is a plain markdown template with sections and inline comments. Fork the repo, delete the comments, fill in your details. First version takes about two hours. The work is thinking — not formatting.
operator-os/ ├── me.md # Who you are, your role, your priorities ├── company.md # Your company, product, strategy, numbers ├── style.md # How you communicate ├── decisions.md # Current open decisions └── people.md # Key stakeholders and relationships
Your role, your scope of ownership, what you delegate, what you refuse, and how you think about your current quarter. The most important file. If your AI knows only one thing about you, it should be this.
Your company's business model, key products, revenue and unit economics if relevant, competitive position, and where you are in the strategy cycle. Keep numbers approximate — the AI needs order of magnitude, not audit-level precision.
Your writing style, email rules, preferred formats, what you always say no to, and 3–5 actual examples of your writing. The style.md file is why AI drafts stop sounding generic. Include real examples — they matter more than any instruction.
Current open decisions with their context: what you're deciding, what you already know, who's involved, and the deadline. Updated weekly. This is the file that makes your AI actually useful for strategic work — not just execution tasks.
Everyone you interact with regularly: name, role, your relationship, and relevant context. The more detail you add, the better your meeting prep and communication drafts become. Add a note every time your perception of a relationship changes.
# me.md ## Role [Your title] [Company name] · [Location] ## What I own - [Responsibility 1] - [Responsibility 2] - [Responsibility 3] ## What I delegate - [What you push down] - [What others handle] ## What I always say no to - [Hard no 1] - [Hard no 2] ## Current quarter priorities 1. [Priority 1 — concrete, not vague] 2. [Priority 2] 3. [Priority 3] ## How I think [2-3 sentences on your operating philosophy. Be honest — this is for your AI, not for LinkedIn.]
Setting up your AI
Three options depending on how much friction you want to remove.
Option A — Claude Projects (5 minutes)
Go to claude.ai → New Project → Project Knowledge. Paste the contents of all five files. Add a system prompt: “You are my AI executive assistant. You have full context about me and my company in your knowledge. Start every response aware of this context.” Every conversation inside this Project arrives pre-loaded with your context.
Option B — Cursor Rules (for builders)
Drop your context files into .cursor/rules/ in any project. Cursor reads them automatically for every AI interaction in that workspace. Add a CONTEXT.md at the project root that points Cursor to your five files.
Option C — System prompt for any AI (universal)
Every major AI tool accepts a system prompt or “custom instructions.” Paste a compressed version of your context files there. Works in ChatGPT, Gemini, Perplexity, and any API-based tool. This is the lowest-friction option if you use multiple models.
I am [your name], [your title] at [company]. ROLE: [2 sentences on what you own] COMPANY: [2 sentences on what the company does, scale, and stage] VOICE: [2 sentences on how you communicate] PRIORITIES THIS QUARTER: [3 bullet points] DO NOT: [What you never want the AI to do — hedge, use filler phrases, etc.] Always respond with my priorities, context, and communication style in mind. Be direct. Be concrete. If you're uncertain, say so.
Getting started
How long does the initial setup take?
me.md takes 30 minutes if you're honest. style.md takes another 30. The other three files are faster — most of the content already exists somewhere. Budget two hours for a first version that's actually useful.
How often do I update the files?
decisions.md weekly — it goes stale fast. me.md and priorities when something structural changes (new role, new quarter, major strategic shift). company.md and people.md as events warrant. The files should reflect how you actually work, not how you worked six months ago.
Do I need to use all five files?
No. Start with me.md and style.md — that alone eliminates most of the re-briefing overhead. Add the others as you run into friction. people.md pays off fastest once you're prepping for meetings.
What if I use multiple AI tools?
The files work in all of them. Keep one canonical version in your context repo and sync it to wherever you need it. If you use Claude as your primary tool, use Option A — Projects make context persistent across all sessions automatically.
Can my team use the same context files?
Yes. A shared company.md that everyone pulls from is one of the highest-leverage things a leadership team can do. It ensures every AI interaction across the team starts from the same factual base. Add team-specific files (team.md, process.md) as needed.
What goes in decisions.md exactly?
Every decision you're actively working through: what you're deciding, the options you're considering, what you already know, who has input, and when you need to call it. Remove decisions once they're made. Add a one-sentence outcome note so your AI understands the result without you restating it.
Where to start
Fork the repo. Open me.md. Write the first section — your role and what you own. That alone takes 20 minutes and immediately makes your AI 3× more useful. Do the rest over the next week.
Fork on GitHub ↗