Tooling for people who run a lot of AI agents.
Most "multi-agent" tools assume you're running two or three agents in parallel. We built these because we're running thirty — and the usual advice (be careful, use branches, review diffs) stops working around agent number five. At that point you need actual infrastructure: isolation, locking, orchestration, and a way to see what every agent is doing without opening thirty terminals.
Everything here is open source and MIT-licensed.
Safety layer for parallel Codex/Claude agents in git repos. Each agent gets its own worktree, claims the files it's editing, and can't clobber files another agent has claimed. Auto-merges Oh My Codex, Oh My Claude, OpenSpec, and Caveman into every worktree so every agent starts tuned, not bare.
npm i -g @imdeadpool/guardexTutorial: guardextutorial.com
Multi-identity Codex CLI. Auto-detects codex login and registers the account to a dashboard so you can see every account you've ever authed and switch with one command. Built because managing 30 agents across 30 accounts by hand is impossible.
npm i -g @imdeadpool/codex-account-switcherDashboard UI for GitGuardex. Shows every active agent, worktree status, file locks, account health, and live usage across all Codex accounts. Rust proxy + telemetry on the backend.
- Recodee — orchestrator for running dozens of coding agents with a Rust runtime, shipping soon.
- Better handoff flows for when an agent hits its usage limit mid-task (currently the ugliest part of running at scale).
- More integrations with Oh My Codex / Oh My Claude so agents in Guardex worktrees start even smarter by default.
Issues and PRs welcome on any of the repos above. We read every one.
If you're also building multi-agent tooling, we'd love to compare notes — open an issue or reach out on any of the project repos.