Chat Client That Checks Its AI Brain Into Git

What it does

HyperChat is a TypeScript chat client that connects to various LLM APIs and wraps them in project-level AI agents. Each project gets a .hyperchat/ directory containing agent configs, memory files, and chat logs that can be committed to Git. It exposes these agents through both a React web dashboard for multi-workspace management and a CLI for quick terminal-driven interactions.

The interesting bit

The “AI as Code” pitch is the standout: instead of cloud-managed prompts and state, everything lives in YAML, Markdown, and JSON files inside your repo. The dual-mode architecture—web for visual workspace juggling and CLI for Agent-first scripting—means the same local brain can drive both IDE-adjacent teamwork and CI/CD pipelines.

Key highlights

• Configuration is plain text in .hyperchat/—agents, memory, and MCP services are Git-trackable by design
• Two-front architecture: a React web UI for multi-workspace tabs and a TUI/cli for fast agent invocation
• MCP protocol support lets agents read and write local files, not just chat about them
• Supports @file path references and Markdown-driven agent command shortcuts
• 5-layer environment variable stacking from system defaults down to CLI args

Caveats

• The README is heavy on marketing language and light on concrete MCP tool examples
• Version 2.0 appears to be mid-migration from a hand-written 1.0 codebase, so stability is unclear
• Documentation is primarily in Chinese, which may limit accessibility for English-speaking contributors

Verdict

Developers who want reproducible, team-shared AI context inside their repositories should look here; if you need a polished, cloud-synced consumer chat app, this is not it.