For the complete documentation index, see llms.txt. This page is also available as Markdown.

About the Swan 1.0 archive

What changed

Swan 1.0
Swan 2.0 (current)
Where to read about it

Two provider roles, ECP (ZK proofs) and FCP (Kubernetes tasks)

One Computing Provider that serves AI inference over a WebSocket connection — no public IP, domain or TLS needed

Providers paid mainly from UBI — a daily SWAN allocation sampled across registered hardware

Providers paid per token at a published payout price for every request they serve

Consumers bought compute through task auctions (Orchestrator, Lagrange, Swan Console)

Consumers call an OpenAI-compatible API and pay per token from a credit balance or a Token Plan

Collateral sized by Computing Units (CU × 3,533 SWAN)

Collateral deposited per provider account, see the current rules

Still running a legacy CP?

The ECP/FCP pages below remain the reference for withdrawing collateral and exiting the legacy network. The recommended path for the hardware itself is to install the current computing-provider and join the inference network — the provider guide takes about fifteen minutes.

In this section

Last updated