Tanaos ships local CPU spam model

This is the kind of narrow, task-specific model release that actually solves a real production problem for developers without dragging in a full general-purpose LLM stack. The pitch is strong: cheap local inference, predictable behavior, and a small enough footprint to fit into ordinary backend workflows.

• –At roughly 500MB and 0.1B parameters, it is sized for CPU deployment rather than GPU-heavy infrastructure.
• –The model is positioned as a first-layer spam filter, which makes sense for support inboxes, contact forms, messaging products, and moderation pipelines.
• –Its MIT-licensed Hugging Face release and connection to Tanaos's Artifex toolkit make it more actionable than a vague “AI-powered moderation” announcement.
• –The one wrinkle is that the packaged workflow still asks users to create a free API key, which slightly complicates the otherwise clean 100% local story.