Welcome to the Antfly Blog
A quick hello from James, and what Antfly actually is: self-learning retrieval for agents.
Hi, I'm James, one of the people building Antfly. Welcome to the blog.
I'll keep this short, because honestly the best way to get Antfly is to use it, not to read about it. But you're here, so let me tell you what we're up to.
Most of what's hard about building with AI isn't the model. It's everything around the model: getting the right context in front of it, at the right moment, without gluing together five services to do it. We think that problem finally has a shape. Retrieval is becoming the new durable primitive of agentic software, the way payments and auth became things you adopt instead of build. An agent only believes what it can retrieve. Get retrieval right and it's useful. Get it wrong and it's confidently wrong.
So we built Antfly: retrieval for agents. Search hands a human a page of candidates and trusts the eye scanning it to pick the right one. An agent has no eye doing the picking. Whatever the engine selects, the agent acts on, so the selection itself has to be right. Antfly is the engine that makes that selection, and it gets a little sharper every time you use it. It runs over the data you already have (your Postgres, your S3), so there's nothing to migrate. No vector store bolted to a graph database bolted to an embedding API. Just one thing. And the whole point is that it's easy: your first query takes minutes, and the code you write on day one still runs at scale.
This is where we'll think out loud about all of it:
- what we're shipping, and why
- how we built it (fair warning, the engineering posts get deep)
- what we're learning about retrieval, agents, and context
- the occasional detour into whatever has our attention
Want to start with the engine instead of the words? Standalone mode is free and runs on your own machine:
antfly standalone
The quickstart takes a few minutes. Come say hi on Discord, or open an issue on GitHub.
Talk soon,
James