ZeusHammer
A voice-first AI agent that combines a local rule-based decision layer for routine tasks with cloud language models, three tiers of memory, and integrations across several model providers including Chinese-language ones.
ZeusHammer is an AI agent that pairs a local intent-recognition and skill-matching layer with cloud language models, so it can handle a portion of requests without a model call and route the rest through a cost-tracked, provider-agnostic model router. It supports voice interaction through Whisper speech recognition and Edge text-to-speech with wake-word detection and automatic language detection, plus real-time audio over WebRTC.
Memory is split across a short-term least-recently-used cache, a SQLite-backed long-term store, and working memory, with a background "meditation mode" that analyzes patterns and updates skills when idle. Other components include a workflow engine, a security system described as a circuit breaker, scanning of commands and files against known vulnerabilities, automatic detection of installed command-line tools, a vector memory store built on Chroma, multi-user collaboration rooms, and integrations with several Chinese-language model providers alongside Apple ecosystem hooks.