Like a lot of you, I went all-in on CLI coding agents — Claude, Codex, Gemini. They're incredible one at a time. But the moment I wanted them to work together — one agent scaffolds, another writes tests, a third reviews — I was back to babysitting terminals, copy-pasting output between tasks, and praying two agents didn't stomp on the same files. So I built the orchestration layer I wished existed. AgentFlow turns your CLI agents into pipelines. You define tasks, wire up dependencies, and let them run. What's inside: 🔀 Mix any agents — Claude, Codex, and Gemini in a single pipeline, each task picks its own 🌿 Git worktree isolation — every task runs in its own worktree, so parallel agents never collide 🔗 Output chaining — a task's output flows automatically into the ones that depend on it ✋ Auto or manual gates — let trusted steps run, pause the risky ones for approval ⏸️ Interactive mode — agents can pause to ask a question or request tool approval mid-run 🔁 Routines — recurring scheduled pipelines for the work you run again and again 🔌 MCP server + Telegram/Slack notifications baked in It's for anyone who's tired of running agents one prompt at a time and wants to actually compose them. I'd genuinely love your feedback — what would you orchestrate first? And if you hit a rough edge, tell me here; I'm reading every comment today. 🙏

AgentFlow

Pipeline-based orchestration for multi AI coding agents

Pipeline-based orchestration for multi AI coding agents