How our open-source database got 12,000 stars before its first sale

Mountainside Labs incorporated in March 2023 in Berlin. My cofounder Lina is a query-engine engineer I'd worked with at the time-series database. We raised $5.6M from two infrastructure-focused funds.

For the first twelve months we built only the open-source database. No cloud product. No pricing page. No sales motion. The pitch to investors had been explicit: give us a year of OSS-only and we'll come to you with a community before we come to you with a product.

§The Discord and the GitHub issues

The v0 was open-sourced on GitHub in October 2023. By April 2024 we had 1,200 stars and a Discord with 180 active members. By October 2024 we had 12,400 stars and ~1,800 Discord members. I personally responded to roughly 4,200 GitHub issues and Discord messages over those twelve months. Lina and one other engineer wrote the actual database. The build pipeline ran on GitHub Actions with release artifacts mirrored on Cloudflare R2 so install commands worked from anywhere in the world without rate limits.

The commercial product launched in November 2024 — exactly a year and a half after incorporation. It is a managed cloud service of the OSS database deployed on AWS with Hetzner as a lower-cost option for self-managed customers. We launched at $290/month for a starter cluster, scaling up. We had 38 paying customers within sixty days. By April 2026 we have 280 customers and ~$3.4M ARR.

The hard part of OSS-first wasn't the engineering. It was the year of no revenue.

My lead investor sent me a one-line email in month nine: "How are you feeling about the timeline?" That email kept me up for three nights. The honest answer was that I was nervous. The Discord was growing. The stars were growing. I had no proof yet that any of those people would pay for anything. I'd staked the entire seed round on a hypothesis that wouldn't be tested for another six months.

§The community blowup

I almost capitulated and shipped a half-baked cloud product in month ten. Lina pushed back, hard. She pointed out that every OSS database company we'd ever worked at had compromised the OSS by rushing the cloud, and that those companies were now fighting their own communities. We agreed to give it the full eighteen months we'd promised investors.

The second-hardest moment was a community blowup in May 2024. A well-known engineer published a blog post arguing that our query planner had a fundamental flaw and was misleading users about its real-world performance. The post was technically half-right. We had a known limitation that we'd documented but not advertised. The blog post made it sound like fraud.

I spent three days writing a response that addressed every technical claim in the post in detail, acknowledging where we were wrong, defending where we weren't, and outlining the specific work we'd ship to fix the legitimate issues. I published the response on our blog. The original post's author updated his post the next week to link to ours. Both posts together became a recruitment magnet — three of our best later hires said they applied because of how we'd handled that exchange. The pattern echoed Larder's reversal email when they tried to loosen buyer curation.

The single thing that turned the cloud product from "nice idea" to "real revenue" was a billing model that priced on ingested data volume per hour, not on cluster size. Every competitor priced on cluster size, which meant customers had to plan capacity. Our hourly ingest pricing meant a customer could spike to handle a launch and pay only for the spike.

This sounds like a small commercial detail. It was actually the difference between customers using us for production workloads and customers using us for experimentation. Production workloads are often spiky. Capacity-priced databases penalize spikes. Ingest-priced databases reward them.

As of April 2026 our average customer's bill varies by 3-4x month over month, and customers love it. Net revenue retention is 144%. Ten of our top fifteen customers came directly from the OSS community — engineers who had been Discord regulars for nine to fifteen months before their company became a paying customer. Billing runs on Stripe with Stripe Tax for VAT compliance, since most of our European customers expected the tax line to be correct on the first invoice.

The OSS-first year, in retrospect, wasn't a delay. It was the entire go-to-market.

§Turning stars into revenue

The first $1.2M of ARR came in the eight months after we shipped Mountainside Cloud — a managed hosted version of the open-source database with a single-region SLA, SOC 2 Type II posture, and a VPC peering option. Twelve thousand GitHub stars converted into a paid pipeline of roughly four hundred companies, of which sixty-one closed at an average ACV of $19,500. The conversion rate from stars to paid (0.51%) was below the 1% benchmark we'd budgeted for, but the per-customer LTV was twice what we'd modeled.

We made one decisive hire: Zara, a former Confluent field-engineering lead, who joined as Head of Solutions Engineering in late 2025. She closed the technical-trust gap that our pure-engineering team kept running into — she could sit in a four-hour architecture review with a customer's staff engineer and walk out with a signed POC. We hired our first three solution engineers under her in early 2026. Pipeline conversion from POC to paid jumped from 19% to 54%. The "open source as distribution, cloud as monetization" pattern is not new — Elastic, MongoDB, HashiCorp, Confluent, GitLab all ran a version of it — but the shape was specific enough that I wrote a deeper version of it down in our internal post-mortem.

1. 1
   OSS-first means OSS-first.

   If you're building infrastructure, give the open-source product a year of undivided attention before you ship a commercial product. Customers will smell rushed cloud products and reject them.

2. 2
   The community is the sales team.

   Of our top fifteen paying customers, ten had been in our Discord for over nine months before paying. The conversion happens slowly, in public, in real time.

3. 3
   Respond to community criticism in detail, in writing, fast.

   A well-written technical response to a critique is the highest-trust marketing artifact you can produce.

4. 4
   Price on the metric your customer's spiky workload actually moves.

   Capacity pricing is a relic of on-prem. Ingest, query volume, or stream-time pricing matches how customers actually consume infrastructure.

5. 5
   A founder responding personally to GitHub issues is a moat.

   It does not scale. That's the point. The thing that doesn't scale is what every later, larger competitor cannot replicate.

What this looked like in practice.

The thing nobody tells you about *open-source distribution* is that GitHub stars are a lagging indicator of brand and a leading indicator of nothing else. We track repo *clones* and Docker image pulls instead — those are the metrics that correlate with cloud-conversion intent.

The thing nobody tells you about *the Apache 2.0 vs SSPL choice* is that it is a five-year decision, not a ninety-day one. We chose Apache 2.0 and accepted that AWS could fork us. Two years in, that has not happened, and the trust dividend with our enterprise customers is worth more than the protection an SSPL would have given us.

The thing nobody tells you about *running the cloud* is that operating a database for somebody else is a fundamentally different company than building a database. Our infrastructure team grew from two to seven in eight months and our on-call rotation became the gating constraint on our hiring.

The thing nobody tells you about *contributors becoming employees* is that the conversion rate is roughly 4% over two years, and those hires onboard in days instead of months. We now treat external contributor relationships as a recruiting pipeline with an explicit budget.

The thing nobody tells you about *competitive pressure from cloud forks* is that the threat is rarely from hyperscalers — it is from the next open-source database that ships a better story to the same audience. We watch the Postgres ecosystem, Tigris, Dragonfly, and TiKV more carefully than we watch AWS.