Show HN: Inferencer – Run and deeply control local AI models (macOS release) (inferencer.com)

Inferencer is a new macOS app that runs state-of-the-art AI models entirely on-device and exposes deep, token-level control tools for transparency and steering. Rather than sending data to the cloud, all inference happens locally; the app supports downloading SOTA models, claims high-speed performance (patent-pending inferencing tech), and offers features like auto-load/unload, retention controls, and parental controls. Visually and functionally it surfaces internals — a token inspector shows per-token probabilities, lets you pick alternate tokens to re-generate responses, displays token entropy to flag low-confidence/contested tokens, and supports token exclusions. It also includes prompt prefilling to seed Assistant messages, enforce output formats (JSON/XML), and "unlock gated" responses, plus markdown and advanced LaTeX rendering (code previews soon). For AI/ML practitioners and privacy-conscious users this is significant: it packages interpretability, debugging, and deterministic steering tools into a local workflow, enabling safer exploration of models, prompt engineering, and forensic analysis without cloud telemetry. Key implications include improved model transparency and user control (useful for alignment and safety research), easier format-constrained generation, and potential shifts toward edge inference—balanced by obvious tradeoffs around local compute, model size, and macOS-only availability. Pricing tiers: a free basic tier with limited token/feature access and a $9.99/month Pro tier that unlocks unlimited token probabilities, entropy, exclusions, and prompt prefilling.