004: Venice: Private AI With Verifiable E2EE

059: Why DeAI: Why Self-Host Your AI?

I tried local AI inference on an M4 Max, failed with dense models, then hit 94 tok/s with MoE architecture via MLX. Here is the honest case for self-hosting, and where it actually beats the API path.

058: Why DeAI: What Is DePIN?

DePIN turns idle hardware into coordinated infrastructure using token incentives. It is the physical layer that makes decentralised AI possible. What it actually means, how it connects to AI, and why the framing matters.

004: Venice: Private AI With Verifiable E2EE

Venice is the rare DeAI project that's actually a product people use, with billions of LLM tokens processed daily through private, uncensored inference. Now with verifiable E2EE and TEE privacy modes powered by NEAR and Phala. Built by Erik Voorhees with no VC funding, no presale. We cover the four privacy modes, the E2EE encryption architecture, uncensored models, VVV token economics, and the centralisation trade-offs.

022: Nosana

Nosana is Solana-based GPU compute for AI inference. Freedom Score 61, Returns Score 51. We cover the CI/CD focus, token staking, and whether Solana speed matters for distributed compute.

057: Open-Weight AI Models Compared

A practical comparison of the best open-weight AI models for local and sovereign inference. Llama, Qwen, DeepSeek, Mistral, Gemma and more, with licence clarity and hardware requirements. Updated monthly.