Watchdog
Local control plane for subagents and agent execution graphs
Start new thread

Watchdog - Local control plane for subagents and agent execution graphs

by•
Subagents, agentic loops and execution graphs make parallel work possible, but they also make it easy to lose track of what each agent is currently doing, which model and reasoning effort it used, amount of tokens being consumed, and other useful info. Watchdog gives you one live view of what your coding agents are doing, what they cost and how they relate to each other. Stop, steer and control your all agents from one place. Watchdog has 3 surfaces- Yard (pixel art UI), Operator, and a TUI.

Add a comment

Replies

Best
Maker
📌
I noticed that people on X (Twitter) were in favour of subagents overall, but were not happy with how using subagents, loops, and execution graphs resulted in agent observability taking a massive hit. Sometimes, even when asked to spawn with cheaper models and lower reasoning efforts, subagents would spawn with expensive high reasoning models and end up burning a hole in the user's pocket. I have faced similar issues too, so I thought it would be a good idea to tackle this problem. My vision for Watchdog is to make it a harness-agnostic control plane. It already supports Codex CLI and Pi, and I plan to add support for Claude Code and OpenCode soon.