Gangline puts Claude Code and Codex agents in a tmux team. Each keeps its own terminal, tools, permissions, and history. Hand a result to a lead, let it recruit teammates by name, and inspect or steer any agent from your terminal.
Install Gangline, then start a team in your repository:
curl -fsSL https://raw.githubusercontent.com/adambiggs/gangline/main/install.sh | sh
gang up -c claudeOr keep the team off your host: with the Docker Sandboxes
CLI signed in, store the Codex login, then run the team's kit with both
harnesses in a microVM over your repository. In the shell it opens, sign
Claude in once with claude and /login, then gang up -c claude.
sbx secret set openai --oauth
sbx run docker.io/adambiggs/gangline --kit docker.io/sbx/claude-mixin --kit docker.io/sbx/codex-mixin .Then tell the lead what you want:
Ask a Codex worker to find how this repository runs its tests. Have it send you the command and supporting file paths, then summarize the answer for me.
The lead briefs a worker, receives its report, and brings the result back. Watch their tmux panes in separate windows or a shared split layout.
The Claude Code lead asks a Codex worker to finish an animated greeting. The worker edits and runs it, reports back, and the finished animation plays. Read the demonstration transcript.
Gangline provides the coordination mechanics: named agents, hitching (launching an agent), collars (files that define harness integration), attributed message delivery, a startup contract, lifecycle logs, and context compaction. Your instructions define the work, who decides what gets built, and how review and acceptance run; Gangline does not manage tasks or claims.
Start with your first team, which also covers installation requirements. Then use the guides for daily tasks, concepts for the mental model, and reference for commands and settings. Troubleshooting starts from what you can see; internals explains the implementation and design choices.
Gangline is licensed under Apache-2.0. See contributing and security reporting.