Agent Panel

An observability platform for autonomous AI agents. See what your agents are doing, track costs and token usage, and get notified when they finish. Includes an open-source Agent Server, a macOS menu bar app, and native iOS companion.

Agent Panel is a monitoring and observability dashboard for autonomous AI agents. The platform has three components: an Agent Server that runs agents on schedules using Claude Code, a macOS menu bar app for persistent operation, and the Agent Panel dashboard for real-time monitoring across all your agents.

How it works

Define agents as YAML or Markdown files with cron schedules, tool permissions, and prompts. The Agent Server discovers them, evaluates schedules, and runs them via the Claude Agent SDK. Every run streams telemetry to Agent Panel, where you can watch agents work in real time, review run history, and track costs.

Agents can search Slack, query Linear, create Notion pages, browse the web with Playwright, and use any MCP server you have configured. The permission system enforces least-privilege access with glob-based allow and deny lists.

Zero-config agent provisioning

You don't create agents in the dashboard. You don't fill out forms. Create an API key, set two environment variables, and every agent run appears automatically. The Panel creates agent records on the fly from the first status event. Agent definitions stay in your YAML files where they belong.

Real-time observability

The live dashboard shows activity cards for every agent: status, elapsed time, turn count, heartbeat status. Click into a run to see the full activity timeline, log viewer with level filtering, and auto-detected output links. When an agent creates a Notion page or files a Linear issue, the output tab shows a direct link with a branded icon.

Heartbeat monitoring catches hung agents. If a run stops sending heartbeats for 90 seconds, it gets marked as failed. No phantom "running" agents in your dashboard.

Agent Server

The Agent Server is a Node.js process written in TypeScript. It handles agent discovery, cron scheduling, file locking, execution via the Claude Agent SDK, and telemetry reporting. Agents support cron schedules, file watch triggers, agent chaining (on_complete/on_failure), interactive workflows via Telegram, and sleep/wake catch-up for missed schedules.

macOS menu bar app

The menu bar app keeps the Agent Server running persistently. It shows active run status, provides a full agent editor with syntax highlighting, and displays run history with detailed execution data. Pure Swift, SwiftUI, zero third-party dependencies. WebSocket connection to the server for real-time updates, with HTTP polling fallback.

iOS companion

The native iOS app provides full feature parity with the web dashboard. Live agent activity, run details with logs and results, API key management, and GitHub OAuth sign-in. Built with SwiftUI, zero external dependencies.

Open source

The Agent Server and macOS app are at github.com/coolasspuppy/agent-server. The Agent Panel dashboard is at github.com/coolasspuppy/agent-panel.