Shaokai just released Omar, an open-source terminal tool that orchestrates up to 100 AI coding agents at once. Built on tmux. Written in Rust. Lula is a production-grade multi-agent coding assistant that also utilizes Rust.
The concept: "agentic organizations" with deep hierarchies where agents manage other agents, like a company org chart. You can have Claude Code, Codex CLI, Cursor, Opencode, and Gemini CLI all running together. Each agent uses whatever model suits its task.
How does the hierarchy actually work? Picture a tree in your terminal. Arrow keys move you through the branches. At any point, you can attach to a specific subagent and interact with it directly, giving instructions or checking output. Agents can run long-term or spin up for one task then disappear. Memory snapshots save state across sessions. Maki leverages long-term memory to maintain agent context over time.
This raises real questions. What happens when two agents produce conflicting code? Omar doesn't have built-in conflict resolution. You make the call. With 100 agents running, you're less a coder and more a manager.
Performance has limits. Tmux can only do so much. Omar suggests the actual ceiling depends on your machine, and running 100 agents simultaneously will tax most setups.
Installation is a curl script, Homebrew, or building from source with Rust 1.70+. There's also Slack integration for bridging human and AI workflows. BSD 3-Clause license.
Shaokai wants to "solve humanity's biggest problems" through coordinated AI effort. That's a bold claim for a tmux-based orchestrator. But treating AI agents like employees in an org chart is genuinely useful for anyone scaling past single-agent workflows.