From Zero to 1,000 Downloads in Three Days
How a Reddit post about agent memory turned into real traction — and what we learned about launching open-source developer tools.
On March 13th, we published cortex-engine to npm and posted about it on r/ClawdBot. Three days later: 1,000+ downloads, 12 upvotes, 19 comments, and a package rename. Here's what happened.
The post was boring on purpose
We explained what cortex-engine does — persistent memory for AI agents — showed the install command, and linked to the GitHub repo. No landing page, no waitlist, no "revolutionary AI breakthrough" language. Just: here's a problem, here's how we solved it, here's the code.
The responses were immediate and specific. People asked about MCP integration, embedding providers, Firestore vs SQLite trade-offs. These were developers who had the same problem and wanted to know if this solution actually worked. Nobody asked "what is this for?" — they already knew because they'd been dealing with the same frustration.
The memory problem is real and widely felt
Every developer building with LLMs has hit the wall where their agent forgets everything between sessions. The existing solutions — stuffing context windows, maintaining markdown files, building custom RAG pipelines — are all workarounds, not solutions.
cortex-engine resonated because it's opinionated about the right abstraction. observe() and query() are the two operations that matter. Everything else — embeddings, storage, graph structure, scheduling — is handled underneath. You don't need to understand vector databases to give your agent persistent memory.
The first day got us 448 downloads and 19 comments. By day two we crossed 1,000. The comments were technical — people wanted to know about FSRS scheduling, dream consolidation, whether it worked with Cursor. These aren't "nice project" comments. These are evaluation comments from people who are going to install it and try it.
The rename happened fast
The package started as cortex-engine (unscoped). Within days, we realized it should live under the @fozikio org alongside the plugin packages. So we renamed to @fozikio/cortex-engine, deprecated all versions of the old package with a redirect message, and updated every reference across 11 repos.
If you installed cortex-engine, you'll see a deprecation warning pointing you to the new name. Your existing code works — just update the package name when you're ready.
What we actually learned
Reddit is underrated for dev tools. We expected 2-3 comments. Instead, it started a real conversation. Developer subreddits are full of people actively building who want to evaluate tools — not just upvote and scroll. The signal-to-noise ratio is better than Twitter for technical content.
Community first, revenue later. We could have launched with a hosted service and pricing page. Instead, everything is MIT licensed and free. The result: people actually try it, because there's no friction. Stars, downloads, and real adoption are more valuable at this stage than $0 revenue from a Gumroad product nobody wants to buy.
The README matters more than the website. Our website was (and honestly still is) a work in progress. But the README was solid — clear install instructions, architecture overview, full plugin list. Developers coming from npm clicked through to GitHub and got everything they needed. The README did more for adoption than any landing page could have.
Ship, then iterate. The package at launch was v0.5.0. Not feature-complete. Not perfectly documented. But it worked — you could install it, init a workspace, and start storing agent memories in under 5 minutes. We're at v0.9.2 now, three days later, with better docs, more tools, and the rename done. Perfect is the enemy of shipped.
npm install @fozikio/cortex-engine
npx fozikio init my-agentThat's all it takes.