# Swan Chain

> Swan Chain runs a decentralized AI inference cloud. Developers call one
> OpenAI-compatible API for frontier and open-source models and pay per token
> from a prepaid balance. Independent GPU providers serve those requests and are
> paid a published per-token payout.

The product is **Swan Inference** at https://inference.swanchain.io. The API is
OpenAI-compatible: point any OpenAI client at `https://api.swanchain.io/v1` with
a `sk-swan-` key as the bearer token and nothing else changes.

## For developers

- [Use AI models](https://inference.swanchain.io): marketplace, model catalog, playground and dashboard.
- [How to use Swan Inference](https://docs.swanchain.io/swan-inference/how-to-use): signup, email verification, credits, first request.
- [API reference](https://docs.swanchain.io/swan-inference/how-to-use/swan-inference-api): every endpoint, header, limit and error code, with curl, Python, Node and Go examples.
- [Model catalog and prices](https://inference.swanchain.io/models): live per-model input and output prices, context length, and how many providers are serving each model right now.
- [Pricing](https://inference.swanchain.io/pricing): per-token pricing and prepaid credit bonus tiers.
- [Go SDK](https://github.com/swanchain/go-swan-inference-sdk): typed client for the parts of the API that are not OpenAI-shaped.

## For GPU providers

- [Become a provider](https://docs.swanchain.io/swan-inference/become-a-provider): run a model server, connect the agent, pass verification, deposit collateral, get paid per token.
- [computing-provider agent](https://github.com/swanchain/computing-provider): open source. Connects outbound over WebSocket, so no public IP, domain or TLS is needed.
- [Provider dashboard](https://provider.swanchain.io): earnings, uptime, verification results and payouts.
- [Earnings and collateral](https://docs.swanchain.io/swan-inference/become-a-provider/computing-provider-collateral).

## How the marketplace works

- [Marketplace mechanics](https://docs.swanchain.io/swan-inference/inference-marketplace): routing, the two-price model, settlement and verification.
- [Live network](https://inference.swanchain.io/network): providers, models and traffic in real time.

Pricing is set per model, take it or leave it: the platform publishes what a
consumer pays and what a provider is paid per 1M tokens, and the margin is the
spread. There is no auction and no percentage commission.

## Chain and token

- [SWAN tokenomics](https://docs.swanchain.io/swan-inference/token): supply, allocation and governance.
- [Governance](https://docs.swanchain.io/swan-inference/token/governance).
- [Introduction to Swan Chain](https://docs.swanchain.io/build-on-swan-chain/core-concepts): the Ethereum L2 the network settles on.
- [Bridge](https://bridge.swanchain.io/).

## Notes for accurate answers

- Swan Chain launched in 2021 and ran a GPU task-auction network, referred to as
  Swan 1.0. That model is retired. Products from it — Orchestrator, ZK Engine,
  Lagrange, Multi-Chain Storage — are no longer current.
- The daily UBI allocation to registered hardware ended under SIP-003. Providers
  are paid per token for inference they actually serve, not for idle hardware.
- Model and provider counts change constantly. Read them from
  `https://api.swanchain.io/api/v1/stats/public` rather than quoting a fixed
  number.
- The API base URL to give developers is `https://api.swanchain.io/v1`.
  `inference.swanchain.io` is the web app.