Ask most founders how they feel about their CRM and you'll get a sigh, usually followed by a story about copying the same customer details between four different tools.
Ben, a software engineer from Mannheim, Germany, spent more than six years building B2B software, most recently as tech lead and team lead, and he saw the same pattern in every company he worked at: a CRM, a project tool, a ticket system, a wiki and an approval flow, all stitched together by people copying data between them.
His answer is Customermates, an open-source workspace that started out as a CRM and is growing into what he calls an agent operating system, where a company's AI agents and its people work on the same customers, knowledge and rules. It's bootstrapped, much of it was written alongside Claude Code, and it launched publicly on November 1, 2026.
Ben runs his own sales and marketing on it, with agents doing the work and him approving whatever goes out, and he says the repo has passed 200 GitHub stars ahead of an open-source launch week starting November 16.
In this Q&A he explains the four-pillar idea behind the product, why personal one-to-one messages have beaten every automated outreach tool he's tried, and what he'd do differently if he started again.
[Screenshot placeholder: Customermates workspace showing a record list with the Mate agent panel open]
1. Who are you, and what are you building?
I'm Ben, a software engineer from Mannheim, Germany. I studied computer science at Heidelberg University and spent more than six years building B2B software, latterly as tech lead and team lead. I'm building Customermates, an open-source agent operating system: one workspace where a company's AI agents and its people work together on the same customers, knowledge and rules. The CRM is where we started, not where it ends.
2. Where did the idea come from, and what tipped you into actually building it?
Every company I worked in, from corporate settings to my own projects, ran the same stack in different disguises: a CRM, a project tool, a ticket system, a wiki, an approval flow, and people copying data between them. I realized almost all of it comes down to four pillars:
- Records: customers, deals, projects, whatever the business tracks, in lists you shape yourself.
- Knowledge: how the company works, written down so people and agents can read it.
- Rules: who may see, change, send and approve what.
- Agents: the ones that do the work, such as reading the inbox, updating records, drafting the follow-up, and asking a human when it matters.
Once AI agents got good enough to work inside such a system, every tool I had used could be rebuilt on those four pillars, configured rather than coded. That was the moment I quit my job at the start of 2026 and went all in, bootstrapped.
3. What did shipping the first version actually take?
I went full-time in June 2026, ran a beta from March and launched publicly on November 1, 2026. It's built with TypeScript, Next.js, PostgreSQL and Docker, much of it written alongside Claude Code. No outside money; the main costs are hosting and AI tooling. The best proof is that we run our own company on it: our sales and marketing are done by agents working inside Customermates, with me approving what goes out.
4. How did you get your first users, and what has moved the needle since?
LinkedIn, Reddit, Indie Hackers and directory listings, with a lot of personal one-to-one messages written as a founder about something the person had actually posted. That has worked far better than any automated outreach. Being open source helps too: developers can read the code and self-host it before they trust it.
[Screenshot placeholder: Customermates GitHub repository page showing the AGPL licence and star count]
5. How does it make money, and where are you now?
Self-hosting is free under the AGPL. The managed EU cloud has paid plans (Starter, Pro, Business, Max) that add Mate, our built-in agent, and the connected inbox. We're at the very beginning: the first small teams are on the cloud, and the focus now is turning conversations into long-term customers. The repo has more than 200 GitHub stars.
6. What has been the hardest part, and what would you do differently?
Explaining it. "CRM" makes people think of forms they hate filling in, and "agents" makes them think of hype. The breakthrough in conversations is showing it: describe your business in a few sentences and watch the workspace build itself. If I started again, I'd lead with that from day one, and set up the first customers by hand with them.
7. Which founders do you watch or learn from, and what is next for Customermates?
I learn most from founders who build in public, especially in open source, where you can see the decisions and not just the results. Next is our open-source launch week from November 16, and partnerships with agencies that want to build their clients' tools on the same four pillars instead of stitching SaaS together.
What to take from Ben's story
Two ideas are worth stealing even if you never touch a CRM. The first is Ben's habit of writing to people about something they actually posted, which backs up what we heard from Chukwuezugo Aronu of Build n Booked on why a targeted list beats a loose one. The second is his advice to show the product doing the thing in the first conversation instead of explaining the category. If you're weighing how far to lean on automation in your own outreach, our guide to the best cold email tools for founders is a useful companion, and our Lindy AI review looks at another way of putting agents to work.
Want to try Customermates? You can find it at customermates.com, and the open-source code is on GitHub.
Have a story of your own? Share your founder story and we'll feature it.
