Herdr Let Codex Drive Claude Code and a Local Model

Companion setup guide: https://productiongrade.substack.com/... Starter configuration and templates: https://github.com/Production-Grade/h... Herdr gives the coding agents you already use a shared terminal workspace. It keeps their sessions available, shows which process needs attention, and lets you prompt, read, wait for or interrupt an agent from another pane. In this walkthrough, Codex coordinates a feature change, Pi runs a local Qwen model for implementation, and Claude Code reviews the design and code. The work moves from design review into a written implementation brief, then back through code review and a follow-up. The point of the test is Herdr itself: whether it works as a practical control plane for multi-agent development. The local Qwen deployment used an Apple M5 Max with 128 GB of unified memory. We also cover readable session names, sidebar and pane controls, scratch terminals, worktrees, status labels, and detach/reattach behaviour. 00:00 A workspace for coding agents 00:22 Sending work between sessions 00:58 The Relayboard experiment 01:23 Requesting the design review 02:13 Carrying the review into a brief 02:40 Sending the implementation to Pi 03:32 Reading the implementation result 04:12 Returning the work to Claude 04:26 Follow-up and final review 04:43 Workspace controls 05:50 Detach and reattach 06:25 Working, done and idle 06:41 Other uses and parallel edits 07:34 Findings and opinions 08:36 What to keep for the next run 09:09 Getting started 09:25 Companion guide and sign-off