Littlesheepxy/zhigeng

What it solves

Zhigeng is a vendor-neutral AI agent designed to act as a personal memory and voice input layer for your computer. It solves the problem of having to repeatedly explain context to different AI models by creating a local, user-controlled long-term memory of your activities, contacts, and habits, which can then be shared with any AI model of your choice.

How it works

The system monitors local activity (windows, conversations, clipboard) to build a "trail" of context. It uses local ASR (Automatic Speech Recognition) engines like SenseVoice or Whisper for voice input, and allows users to provide their own API keys (BYOK) for LLM reasoning.

On macOS, it integrates as a menu bar app that provides structured voice-to-text input and smart replies based on the current screen context. It can also dispatch tasks to local agents like Codex or Claude Code for complex operations.

Who it’s for

Power users on macOS (and eventually iOS) who want a unified, AI-enhanced input method and a personal knowledge base that remains local and independent of any single AI provider.

Highlights

• Context-Aware Voice Input: Converts speech to polished, structured text tailored to the specific app or tone (e.g., email vs. chat) rather than simple transcription.
• Local Memory: Maintains a local record of people, projects, and habits that the user can view, edit, or delete.
• Smart Replies: Generates multiple draft replies based on the current conversation and application context.
• Vendor Neutrality: Supports multiple LLM providers via BYOK (OpenAI, Anthropic, Zhipu, Kimi, etc.) and local ASR engines.
• Agent Orchestration: Can delegate complex tasks to other local AI agents like Codex or Claude Code.