Agent Belay

Let your AI move fast without breaking things.

---

Agent Belay is a **human-in-the-loop gateway** for AI coding agents like [Claude Code](https://docs.anthropic.com/en/docs/claude-code) and [GitHub Copilot](https://github.com/features/copilot). It intercepts tool requests (file edits, shell commands, web fetches, etc.) via hook events, displays them in a review UI, and lets you approve or deny each action before it executes.

     

Key Features

• Approval UI — Review pending tool requests with syntax-highlighted diffs, command previews, and context
• Protection Engine — Built-in rules that block dangerous operations (destructive commands, credential access, supply-chain attacks)
• Risk Analysis — Optional AI-powered risk scoring (1-5) via Claude CLI or GitHub Copilot to auto-approve safe operations
• Always Allow — Grant persistent permissions for trusted tool patterns
• Away Mode — Disable timeouts for remote/async approval workflows
• History — Searchable log of all past approval decisions
• Usage (experimental) — Token usage, cost (USD), and p50/p95 latency per harness, scanned from on-disk session files
• System Tray — Runs in the background with badge notifications f