Hey Product Hunt 👋 Maker here. I built Forge because running multiple coding agents in parallel was driving me a little crazy. I'd have Claude Code in one terminal, Codex in another, OpenCode doing its thing — and they were all editing the same working tree with no shared notion of "what work is in flight" or "who owns what." I lost diffs. I lost track of which agent broke main. It was chaos. I tried a handful of tools in this space and none of them quite fit how I work — some were closer than others, but I kept hitting walls. So I built the thing I wanted. The core idea: **agents shouldn't edit your repo directly.** Each task gets its own isolated git worktree, runs through a CI gate you define, and lands in a review step before anything merges. Forge coordinates all of it. What's in the box: 🔧 Self-hosted, runs locally, MIT-licensed 🤖 Connects to any MCP-speaking agent — Claude Code, Codex, OpenCode, Cursor 🌳 Isolated git worktrees per task, so agents never stomp on each other ✅ CI gates and human review before merge 🖥 CLI, REST API, and an optional web UI ⚡ `brew install forgeailab/tap/forge` or `docker compose up -d` It's v0.1 — early, moving fast, the loop works end-to-end but the edges are rough. The whole reason I'm posting today is to find out what's broken in *your* multi-agent workflow so I know what to fix next. Would love your honest feedback 🙏

Forge

Self-hosted workflow engine for all favor agents

Self-hosted workflow engine for all favor agents