Decentralised AI. Compute marketplaces, model coordination, on-chain inference — and the bet that AI's most consequential layer should not be owned by three companies.

Why does DeAI matter?

DeAI is part of FRQNCY's Money domain in the Fund pillar — a curated map of what to read, who to follow, and how the topic connects to consciousness, capital, and the network state.

What are the best resources to learn about DeAI?

FRQNCY curates resources for DeAI at https://frqncy.network/deai/.

DeAI — FRQNCY Network

Take a Course

Intermediate — coders new to ML Introduction to Machine Learning for Coders 12 lessons · ≈ 24 hours · 12 lessons · Jeremy Howard · fast.ai · 24 hours of video →

Curated Resources

5 resources · 5 orgs

org

DeSciWorld

Aggregator and community hub for the DeSci ecosystem since 2021. Maintains a directory of DeSci projects, an events calendar (in-person and online), a jobs board, and the upcoming science.tech tooling stack + P2P funding marketplace. Runs the flagship World of DeSci conference series and seeds local chapters across four continents — DeSci Asia, Berlin, Boston, China, India, Japan, Korea, LATAM, London, Mexico, Montreal, Prague, Tokyo, Africa, IberoAM. The 'one-stop shop' for finding what's happening in DeSci globally.

Visit →

org

Kaggle

The world's largest community of machine-learning practitioners. Runs the canonical ML competitions where the line between hobbyist and world-class is decided by a leaderboard, not a CV. Hosts hundreds of thousands of public datasets and notebooks — the closest thing the field has to a shared workshop. Free tier includes GPU/TPU compute. Where a generation of ML engineers learned the craft in public. Owned by Google since 2017 but the community culture predates the acquisition by a decade.

Visit →

org

IronClaw (NearAI)

Open-source personal AI agent — Rust, security-first, defense-in-depth. A re-implementation of OpenClaw rewritten for privacy: WASM sandbox for untrusted tools, capability-based permissions, endpoint allow-lists, credential injection at the host boundary so secrets never touch tool code, leak detection on requests and responses, and prompt-injection detection layered on top. All data stays local (Postgres + pgvector), no telemetry. Multi-channel: REPL, HTTP, Telegram, Slack, web gateway. Apache/MIT. ~8.5k stars. Maintained by Near AI.

Visit →

org

OpenClaw

The original open-source personal-AI agent project IronClaw is a re-implementation of. TypeScript, Docker-sandboxed tool execution, self-expanding via dynamically built tools, MCP-protocol client. Designed around the same principle: an assistant that works for you, runs locally, doesn't phone home. Where the architecture for the whole 'sovereign personal agent' pattern got worked out.

Visit →

org

Hermes (Nous Research)

Open-weights LLM family from Nous Research — the most-shipped uncensored model line in the open ecosystem. Hermes 4 (including the 405B variant) trades the safety-tuned conservatism of frontier closed models for a tool-use- and agentic-workflow-tuned generalist. Available via OpenRouter, Together, Fireworks, and locally via Ollama. The default 'open alternative' when a workload needs an LLM that won't refuse legitimate requests or leak its system prompt at the slightest pressure.

Visit →

Explore

Topic · 4 shared