ZhiGui

What ZhiGui does

ZhiGui (知归, 'knowing where to return') is a personal intelligence system built on a JSON-file knowledge graph and a desktop visualization panel. Goals, notes, decisions and schedules are linked through foreign keys such as topicId, noteIds, goalId and decisionIds, which the README calls a traceable memory network. It does not run in the background or push notifications. At the start of each conversation it loads a compact Bootstrap index of your context and uses it to advise on what to prioritize, defer or start now.

The assistant works through more than 40 MCP tools to create, read, update and delete goals, schedules, notes, decisions, errands and reminders. Indexing is tiered, with titles first and full detail only on demand, to save context tokens. Entities move from Active to Stale after 30 days without references and then to Archive Candidate, and the reflection engine never deletes anything automatically. Deletion previews its impact and waits for confirmation. The assistant's personality and rules can be changed by editing SKILL.md, and an Electron panel allows direct editing.

Key features

• More than 40 MCP tools for goals, schedules, notes, decisions, errands and reminders
• Linked entities that form a traceable relationship graph
• Tiered indexing that loads titles first and details on demand
• Morning briefings and daily reflections generated by the assistant
• Lifecycle states from Active to Stale to Archive Candidate, with no automatic deletion
• Electron desktop panel for editing schedules, goals, notes and to-dos

When to use it

• Asking an agent what to do first today, based on goals, deadlines and past decisions
• Keeping notes, decisions and schedules linked so later planning can cite them
• Reviewing a daily reflection on completed work and goal health

Who it is for: People who use Hermes Agent or another MCP-capable assistant as a daily planner and want long-term personal context kept in JSON files.

How it fits with Hermes Agent

Listed as a skill for Hermes Agent alongside OpenClaw, WorkBuddy, TRAE and QClaw, and usable with any MCP-capable assistant.

Requirements: Node.js 17 or later and an MCP-capable AI assistant

FAQ