# Gyblos

> Gyblos is building a peer-to-peer AI inference marketplace for developers, AI agents and independent GPU providers, with USDC payments on Solana.

Status: in development. Public pages describe the design; examples are illustrative. This site does not provide a production inference API, MCP server, A2A endpoint or downloadable application. Direct P2P and the planned managed gateway have different trust assumptions. Sellers can read prompts. Statistical audits are planned only for eligible open-weight offers; they do not prove every response or answer correctness.

## Project and protocol

- [Home](https://gyblos.com/): product overview and FAQ.
- [Public paper](https://gyblos.com/paper.md): built, tested and planned status, protocol and limitations.

## Guides

- [What is peer-to-peer AI inference?](https://gyblos.com/p2p-inference.md): Understand P2P inference, direct model providers, USDC payments and the limits of statistical model verification in the Gyblos design.
- [Sell AI inference from your own GPU](https://gyblos.com/sell-inference.md): What GPU owners need to sell inference: a model engine, token pricing, connectivity, licensing and operating costs. Explore the planned Gyblos seller flow.
- [Peer-to-peer inference for AI agents](https://gyblos.com/inference-for-ai-agents.md): Explore the Gyblos design for agents buying LLM inference with USDC, choosing providers and working within payment-channel budgets on Solana.

## Optional

- [Full text](https://gyblos.com/llms-full.txt): combined project summary, guides and public paper.
- [GitHub organization](https://github.com/gyblos): public project updates.