# SolanaLM

> SolanaLM is a production-ready hybrid decentralized AI network combining large language model inference and federated learning, with settlement and node incentives on Solana. Operators earn SOL by serving inference requests and contributing GPU cycles to training rounds.

SolanaLM bundles a FastAPI gateway, a Python client SDK with OpenAI compatibility, and a federation runtime that implements FedAvg, FedProx, FedAdam, and SCAFFOLD. Inference nodes can run PyTorch/Transformers, llama.cpp, or proxy to OpenAI, Anthropic, Cohere, and Ollama. Privacy-preserving onion-routed inference and differential-privacy aggregation are first-class. MIT-licensed, Python 3.12+.

## Docs

- [Home](https://solanalm.cryptuon.com/): hero, pitch, and feature overview for the network
- [Features](https://solanalm.cryptuon.com/features/): full feature breakdown across inference, FL, privacy, settlement, observability
- [Architecture](https://solanalm.cryptuon.com/architecture/): gateway, registry, nodes, privacy layer, settlement, and the request lifecycle
- [Quickstart](https://solanalm.cryptuon.com/quickstart/): clone, poetry install, configure a node, first OpenAI-compatible request
- [FAQ](https://solanalm.cryptuon.com/faq/): wallets, models, privacy, costs, settlement, and federated learning answered
- [Glossary](https://solanalm.cryptuon.com/glossary/): key SolanaLM, inference, and Solana settlement terms defined
- [About](https://solanalm.cryptuon.com/about/): project mission, thesis, and the team behind SolanaLM
- [Contact](https://solanalm.cryptuon.com/contact/): talk to the team

## Use cases

- [Use cases index](https://solanalm.cryptuon.com/use-cases/): overview of what people build on SolanaLM
- [Privacy-preserving inference](https://solanalm.cryptuon.com/use-cases/private-inference/): onion-routed, confidential LLM inference where the operator can't read your prompt
- [Monetize idle GPUs](https://solanalm.cryptuon.com/use-cases/monetize-idle-gpus/): earn SOL by serving inference and training rounds on spare DePIN hardware
- [Federated model training](https://solanalm.cryptuon.com/use-cases/federated-training/): coordinate FedAvg/FedProx/FedAdam/SCAFFOLD rounds across distributed nodes
- [OpenAI drop-in replacement](https://solanalm.cryptuon.com/use-cases/openai-drop-in-replacement/): switch existing apps to SolanaLM with a base-URL change

## Comparisons

- [Comparisons index](https://solanalm.cryptuon.com/compare/): how SolanaLM differs from other decentralized AI networks
- [vs Bittensor](https://solanalm.cryptuon.com/compare/bittensor/): Solana settlement and OpenAI schema vs Polkadot subnets
- [vs Akash Network](https://solanalm.cryptuon.com/compare/akash/): purpose-built LLM serving and FL vs generic compute leasing
- [vs Together AI](https://solanalm.cryptuon.com/compare/together-ai/): open-source, self-hostable, privacy circuits vs hosted provider lock-in
- [vs Ollama](https://solanalm.cryptuon.com/compare/ollama/): multi-node, monetized, federated vs single-host local runner

## Blog

- [Blog index](https://solanalm.cryptuon.com/blog/): deep dives on federated learning, Solana micropayments, and OpenAI-compatible gateways
- [Why we picked Solana for per-request AI inference payments](https://solanalm.cryptuon.com/blog/why-solana-for-ai-inference-payments/): the engineering case for ~0.4s finality and ~$0.00025 fees in an inference market
- [OpenAI API compatibility is a distribution strategy](https://solanalm.cryptuon.com/blog/openai-compatibility-as-distribution-strategy/): why speaking OpenAI's schema is the cheapest way for existing apps to adopt the network
- [FedAvg, FedProx, FedAdam, SCAFFOLD: what breaks in production FL](https://solanalm.cryptuon.com/blog/federated-learning-algorithms-in-practice/): what each federated-learning algorithm is for and how to choose
- [The threat model for onion-routed LLM inference](https://solanalm.cryptuon.com/blog/onion-routed-inference-threat-model/): what attacks the 3-hop onion circuit actually stops, and what it does not
- [SolanaLM vs Bittensor: a deep dive](https://solanalm.cryptuon.com/blog/solanalm-vs-bittensor-deep-dive/): an honest engineering comparison on settlement, schemas, and subnet doctrine
- [Running a SolanaLM node: real numbers from week one](https://solanalm.cryptuon.com/blog/running-a-node-economics-real-numbers/): hardware, hours, and payouts from a week of running an inference node

## Optional

- [RSS feed](https://solanalm.cryptuon.com/rss.xml): blog feed
- [Sitemap](https://solanalm.cryptuon.com/sitemap-index.xml): full URL index