Agent Monitor

Appearance

Quick Start ​

Prerequisites ​

  • Node.js 18+
  • Claude Code CLI (claude) or OpenAI Codex CLI (codex)
  • OpenCLI runtime (@jackwener/opencli) is auto-synced during server install (OpenCLI itself requires Node.js 20+)
  • Git (optional; needed for worktree isolation in git repos)

Installation ​

bash
git clone <repo-url> && cd AgentMonitor
npm install

For local-only use, you do not need relay setup. Install the CLI you want to use, then run the app directly on your machine. When you run cd server && npm install, Agent Monitor also performs a best-effort sync to @jackwener/opencli@latest.

Running ​

bash
# Development mode (hot reload)
npm run dev

# Production build
npm run build
cd server && npm start

The local dev UI is available at http://localhost:5173, and the API server runs on http://localhost:3456.

Creating Your First Agent ​

  1. Navigate to New Agent in the nav bar
  2. Enter a name, working directory, and prompt
  3. Use Browse to pick a directory — if a CLAUDE.md or AGENTS.md exists, you'll be prompted to load it
  4. Select provider (Claude Code or Codex)
  5. Select a model from the runtime-detected dropdown (or keep default)
  6. Configure flags (e.g., --dangerously-skip-permissions, --chrome, --permission-mode)
  7. Click Create Agent

Tip: use the built-in OpenCLI Skill Starter template from Load template... if the task needs website/browser/desktop automation, or Karpathy Coding Guardrails if you want stricter coding guardrails.

If the working directory is a git repo, the agent will start in an isolated worktree branch. Otherwise, it works directly in the target directory.

Agent Monitor keeps instruction content compatible across providers:

  • Claude agents use CLAUDE.md
  • Codex agents use AGENTS.md
  • If only one exists, Create Agent can still load it and write the provider-specific file when the agent is created

Model behavior is provider-specific:

  • Claude applies the selected model via --model at process start
  • Codex applies the selected model by prefixing the first turn with /model <name>

Using the Dashboard ​

The dashboard shows all active agents as cards with:

  • Provider badge (CLAUDE / CODEX)
  • Current status (running, stopped, error, waiting_input)
  • Cost and token usage
  • Latest message preview

Click any card to enter the full chat interface.

Settings ​

Open Settings from the dashboard to configure:

  • Auto-delete retention: Automatically remove stopped internal agents after a configurable period (default 24h, set to 0 to keep forever)
  • Session-file delete policy: Choose how delete handles provider session files (ask every time, do not purge session files, always purge session files)

Language ​

Use the language dropdown in the top-right corner to switch between 7 supported languages: English, 中文, 日本語, 한국어, Español, Français, Deutsch.