ThoxyWeb
Chat with THOXY AI assistant locally in your browser
At THOX.ai, we build local-first, privacy-first AI around the people using it. Our focus is making capable AI practical across edge devices, workstations, portable hardware, and embedded systems—without making cloud dependence the default or compromising user control. We research how models, runtimes, devices, and agents can work together to deliver useful, reliable intelligence closer to the data. Areas of interest * Edge AI and efficient models: Small and specialized language models, custom architectures, instruction tuning, quantization, distillation, and hardware-aware inference optimization. * Portable and privately connected AI: Offline-ready runtimes, model portability, and permissioned networks that connect authorized devices and compute resources. * Intelligent routing and agentic systems: Task-aware model selection, multi-agent coordination, and Digital Employees operating with scoped permissions, reviewable actions, and human oversight. * Multimodal and human-centered experiences: Language, speech, vision, and digital-human interfaces that make advanced AI more accessible and useful in everyday work. Across ThoxOS™, MeshStack™, and the upcoming ThoxWork™, our goal is to turn these capabilities into a coherent experience—not a collection of disconnected tools. We value reproducible evaluations, transparent model provenance, honest hardware compatibility claims, and explicit user choice about where data is processed. Your AI. Your Data. Your Rules.™
Chat with THOXY AI assistant locally in your browser
Permanent OpenAI-compatible upstream for llm.thox.ai
Your AI. Your Data. Your Rules.™ Local-first intelligence.
Generate text responses with ThoxAir offload model
ThoxIntel-27B - THOX flagship reasoning model by Thox.ai
OpenAI-compatible THOX node in the MeshStack mesh
Chat with an uncensored AI assistant for answers and live code
BitNet b1.58 ternary LMs small enough for a microcontroller
Chat with ThoxMini-3B for concise factual answers
125M from-scratch micro LM. Local-first demo.
0.5B on-device edge chat model. Local-first demo.
9M BitNet role model for edge. Local-first demo.
16M BitNet role model for edge. Local-first demo.
9M BitNet role model for edge. Local-first demo.
Preference router LoRA on Qwen2.5-1.5B-Instruct. Local-first
327M from-scratch THOX decoder. Local-first demo.
OpenAI-compatible image generation on ZeroGPU
Keyless multi-provider web search in the SearXNG JSON shape
OpenAI-compatible speech-to-text on faster-whisper
Chat with AI for answers and Rust code help
OpenAI-compatible coding backend for THOX agents
Calculate cosine similarity between two sentences
Registry and router for THOX mesh nodes