I run a payments company from Telegram.
This is the AI team behind it. Free. Open source. You approve every step.
By Panche Isajeski · Tallinn · open source, MIT
Developers: install it from GitHub →
- Free and open source: the roles, steps, gates and templates are on GitHub under MIT.
- For SaaS founders, agencies and freelancers.
- Six departments: strategy, finance, product and engineering, growth, sales and customers, operations. A chief of staff runs them and asks you at every step.
- They draft and propose. You approve. Nothing goes out without you.
- Free and complete on its own. Add your own knowledge, or as a member plug in mine: grant programs open now, 60+ plays for first customers, free places to list your product, and how I got named in ChatGPT and Google answers.
Install it in two lines
In a terminal with Claude Code installed:
claude plugin marketplace add PancheI/BootstrappedOS
claude plugin install bootstrapped-os@bootstrapped-osThen open your project, type /day0 and answer five questions. Step by step, with what you should see: the install guide.
See it run: the live workshop
Live with 30+ founders: how I run my company with these agents, with real numbers. Four chapters, with the real numbers:
The phone ops bot in chapter 1 is my own private setup, not part of the OS.
Free and complete. Add your knowledge, or plug in mine.
- Built in, free: every role comes with proven steps. Install it and go.
- Yours, free: add your own notes, rules and numbers. Every agent follows them first.
- Mine, for members: my knowledge on tap, inside your agents, updated every week:
- Get found on Google and ChatGPT: my site’s Google trust score went 5 → 33 in one month (6×), and AI named my product in 4 of 9 buyer questions.
- First customers: 25 of 90 people I messaged replied, 2 sales on launch day. The exact messages and follow-ups.
- Funding: $2M in grants raised for my last startup. The programs open now, with deadlines, and $300 in AI credits on day one.
- AI team: mine costs ~$5 a day and shipped 791 releases in 114 days. The setup and the lessons.
- Mistakes to skip: 629 notes of what went wrong, each with the reason.
It keeps getting better. I run my company every day and keep adding what I learn. Your agents get it automatically, so they get better over time. One key from your account; your own notes always win.
- Without: “Where should I list my launch?” gets a generic list of sites, some dead, most worth nothing.
- With my knowledge: the places I checked myself, how to submit, which links are real, and when I last checked, plus the play I ran around them and what happened.
Your key is in your account the minute you join.
Pick an agent
Start with the one job that hurts most. Each agent drafts and proposes; you decide what goes out.
Finds people asking for what you sell this month, with the link and why they fit.
Youwrite the reply: AI finds them, you talk to themFinds the grants and AI credits you qualify for, with the deadline.
Monthlychecked on the official pagesSends you one brief every morning: what happened, what needs you.
go 1 3you approve by numberPrices your product against similar ones and shows your break-even.
2–3price models with live examplesLists you in the right places and gets you into Google and AI answers.
#2in Google’s AI answer, one month inTells you where users get stuck, with one fix a week.
5numbers every Monday, one changeWrites your post from a real win, in your voice.
1real win per post, no fillerBuilds your landing page: three options, then checked on desktop and phone.
791releases of my own in 114 daysEvery agent is in the free repo. Members add the knowledge each one asks for.
The departments
| Department | Roles | What you get |
|---|---|---|
| Office of the CEO | Chief of Staff | Runs the stages, picks the right model for each job, keeps the knowledge base, asks you at every gate. |
| Strategy | Head of Research, Idea Critic, Head of Positioning, Brand Designer | A niche idea from real buyer complaints, attacked before you see it; why anyone would pay instead of doing nothing or asking a free AI chat; a name and a brand. |
| Finance | CFO, Head of Grants | Two or three price models with live precedents and break-even, placed against market data; payments set up; grants and credits you qualify for. |
| Product & Engineering | Head of Design, Head of Infrastructure, engineering team, Browser Checker | A landing page with your product working on it, deployed, with login and checkout; every page checked in a real browser. |
| Growth | Head of Growth, Head of Content, Creative Director | Getting found (Google, ChatGPT, listing sites, AI answers), posts and lead magnets from real wins, a launch video from your real screens. |
| Sales & Customers | Head of Sales, Customer Writer, Head of Support | Your first paying customers: the people who show the problem this month, follow-ups, then replies to every customer. You write to buyers yourself. |
| Operations | Head of Analytics, Watch | Five numbers every Monday with one change to make, and a daily read-only check of what needs you. |
From idea to first customers, one step at a time
/start-from-zero asks about you and your idea, then runs the departments in order. At the end of each stage it shows the result and waits for your yes: the idea, the positioning, the brand, the price, the page, the deploy, the video, the launch. knowledge/progress.md shows where it is, which model each stage used and how many tokens it took. Already live? /day0 asks your revenue and starts you where you are.
One knowledge base, so they work as a team
It creates a knowledge/ folder in your project: what you know, the idea, the positioning, customers, decisions, numbers and a daily log. Every agent reads it before a task and writes one line after. It is plain Markdown you own.
They draft. You decide.
One rule, enforced: the agents read, draft and propose. Sending, publishing, paying, deploying and deleting are blocked or ask you first. The gates come with tests you can run yourself, in the repo.
What it costs to run
| Setup | What you pay for | Per month |
|---|---|---|
| The same roles as freelancers | Part-time rates for these roles (estimate) | ~$15,000 |
| Maker on a budget | A $20 AI plan + free tiers of GitHub, Vercel, Supabase and PostHog | $20 |
| My setup today | Claude Max; everything else on free tiers or startup credits | $100–200 |
The software is free. Your AI plan is paid to its provider, not to us. The chief of staff sends hard calls to the strongest model and routine work to cheaper ones, so you pay for thinking only where it matters.
What it works with
GitHub, Vercel and Supabase for the product, AgentaOS for payments (my company), PostHog for numbers, Gmail for drafts, a visible Chrome for research and checks, ElevenLabs for the video voice. Built on the open Agent Skills and MCP standards, it runs in your own AI coding agent.
What it can’t do
It doesn’t send, post or apply on its own, and it doesn’t guarantee revenue. It does the research, the drafting and the remembering, so your hours go to the decisions only you can make.
Where it comes from
I run my own company this way: 114 days, 1,033 hours and 3,313 agent runs, while AgentaOS went from $0 for four months to ~$10k MRR. Bootstrapped OS is that system, cleaned up so it works for any small business, and open source. My company-specific skills and my phone ops bot stay private. How I run it, with all the numbers: how to run a one-person company with AI agents. What an open-source AI team is, in plain words: what is an open-source AI team for a one-person business? And the full story.
Questions
What is Bootstrapped OS?
A free, open-source AI team for a one-person business: 21 agents in six departments (strategy, finance, product and engineering, growth, sales and customers, operations), run by a chief of staff. They share one knowledge base and take you from a niche idea to a launched product and your first customers. You approve everything before it goes out. It runs in Claude Code.
Is it free?
Yes. The whole method is on GitHub under the MIT license. Members of Bootstrapped Makers also plug in my knowledge: the plays I ran with their results, grant programs open now, places to list your product, and every lesson. I keep adding to it, so your agents get better over time.
I don’t sell software. Is it for me?
Yes. It works for any one-person business: digital products, ebooks, courses, marketing, consulting or services. The engineering module is optional; research, pricing, payments, getting found and first customers work the same.
Which AI does it run on?
Claude Code, on your own plan. It is built on the open Agent Skills and MCP standards. A $20 plan is enough to start.
Do I need to code?
No. You answer five questions and approve each step. It builds the page and the product with you. Members can set it up with me.
Does it send emails or post for me?
No. It drafts and proposes; sending, posting, paying and deleting wait for you. That is the core rule, and it is tested.
How much does it cost to run?
The software is free. Your AI plan starts at about $20 a month; mine is Claude Max at $100–200. It is paid to its provider, not to us.
What do members get on top?
My knowledge on tap, inside your agents with one key: how my site went 5 → 33 on Google in a month, the messages that got 25 of 90 replies, grant programs open now ($2M raised for my last startup), and every lesson. Plus $300 in AI credits on day one, setup with me and the founders’ chat.