Members

Each member's key is used by both their node and their swarm. A new key, or removal, disconnects their node.

Add member

Tiers

Names swarm asks for in place of a model. Each points at a model some node serves, and says what a second of its work is worth in credits: a bigger model's counts for more.

Set tier

Policy

How the network is used. Nodes and clients say nothing about it; the router decides.

Cache bias How many requests a chat will wait behind to stay on the node that has its context cached. Local bias The same for a member's own node. Also how much more their requests count in the line at it. Keep it below the cache bias. Credit floor Below this balance a member only takes: they are served last, from spare slots. Negative, so newcomers can spend. Luck Of those waiting, the next is drawn by lot, weighted by what each member has given. 1 is the usual; higher evens it out, lower follows credits more strictly. Credit half-life (hours) How long a balance, in credit or in debt, takes to fade by half, so debt is forgiven in time. 0: never.

Save policy

Requests

The latest relayed, with llama.cpp's own timings.

When	From	Node	Model	Path	Status	Time	Tokens	Credits	Prefill	Prompt t/s	Gen t/s