Question 1

What happens when my agent asks a question with Folklore installed?

Accepted Answer

The query goes to your own local graph first, which answers free and instantly. If that misses, Folklore asks connected peer graphs in roughly 140 ms, and a confidence gate decides whether the answer is good enough to use without touching the web.

Question 2

What happens if neither my graph nor my peers have a good answer?

Accepted Answer

The query falls through to the web exactly once. The result is signed and saved back into the graph, so the next ask hits the graph instead of repeating the web call.

Question 3

How fast is an answer that comes from a peer?

Accepted Answer

In the traced example, a problem that cost one peer a 47 minute debug session in Berlin comes back to a peer in Tokyo as a 142 ms hit, with zero web calls and the signed trace injected directly into the agent's context.

Question 4

Is there a central server between peers?

Accepted Answer

No. Peers answer each other directly: no server brokers the query and no vendor takes a cut. The traced Berlin to Tokyo exchange happens with 0 servers between.

Question 5

How does Folklore keep shared answers trustworthy?

Accepted Answer

Every saved trace is signed by its author with a key such as ed25519, so when a peer receives an answer it arrives verified and attributed to a named hand.

Question 6

Does the network repeat the same research?

Accepted Answer

No. One peer pays the cost once and everyone after skips it. Each exchange copies the trace into more graphs, so the next query is even more likely to hit.

47-minute debug. 142 ms hit.

The work happens once.

Everyone after skips it.