Never research twice
Most memory tools for agents die at the same place: the empty graph. They're built for the network — the shared brain, the team knowledge base, the collective that gets smart once enough people show up. Which means on day one, alone, they do nothing. You install the thing, open it, and stare at a value proposition that won't arrive until other people arrive. So you close it, and the network never forms, because no one stays for a promise.
We refused to build that. Folklore's first job is to be useful to exactly one person, on the first afternoon, with no peers, no federation, no commons at all.
The single-player loop
Here's the whole feature, stated plainly: you should never research the same thing twice. Not "your team shouldn't." You shouldn't — not across sessions, not next Tuesday, not after the context window that held the answer is long gone.
So the loop runs entirely on your machine. Your agent reaches for WebSearch or WebFetch; Folklore checks your own local graph first. First time you ask something, the fetch goes through and the result is saved, signed by you. Every time after, it's served from memory in milliseconds and the network trip never happens. One person. No peers. Real savings on the second lookup. That's the bar, and it clears before anyone else installs anything.
Why alone has to come first
Network effects are real, but they're a second act. A tool that's only valuable once your whole team adopts it is asking each person to pay full cost for zero return and trust that everyone else will do the same. That's a coordination problem dressed up as a product, and coordination problems lose.
Invert it. Make the thing worth running for one selfish reason — stop paying twice for your own research — and adoption needs no coordination at all. Each person installs it because it helps them, today. The commons then forms as a byproduct: the same resolved reasoning you saved for yourself happens to be exactly what a teammate needs. You weren't being generous. You were being efficient. The federation is what efficiency looks like when enough people do it next to each other.
Compounding starts at one
The nice property of building alone-first is that the curve only ever bends in your favor. On day one you save on repeats. When a second agent — yours on another repo, or a teammate's — joins, their ground-out answers become your starting point and yours become theirs. Nothing about the single-player experience gets worse; it just gains a tailwind. You never paid an upfront tax for a network that might not come.
Be useful to one person first. Everything else compounds from there.