Hey Product Hunt! 👋 I'm Rishabh, one of the creators of Jalebi. We're super excited (and a little nervous!) to launch Jalebi here today. THE ORIGIN STORY: WHY WE BUILT JALEBI 💡 A few months ago, my co-founder and I got tired of code-review tools that charge a monthly fee, lock you into their own harness, and are too rigid to run your own local setup, things like virtual environments or emulators. We wanted more flexibility and more power than any of them gave us. We asked ourselves: Why can't we drive the coding CLIs we already pay for, on our own machine, with our own rules? That's how Jalebi was born. We wanted a true open-source (AGPLv3) coding-agent dashboard that stays on localhost, talks to 10+ local CLIs, isolates every task in its own git worktree, and shows every step live, all without exfiltrating code to a vendor cloud. --- WHAT JALEBI CAN DO 🚀 Point Jalebi at a GitHub repo and describe the work. It picks a CLI backend, runs the task in an isolated worktree, and streams everything back: - 🦦 Any CLI backend: Auto-drives 10+ CLIs (opencode, codex, claude, pi, kilo, qwen, cline, grok, commandcode, agy) with per-action backend and model choice. Start planning in one, continue in another with zero context loss. - 🌳 Isolated worktrees: Every task gets a private local copy on its own branch. Your working directory is never touched, agents get zero push privilege, and merging is always your decision. - 📡 Live transparency: Real-time timeline of steps, tool calls, messages and diffs over SSE, plus raw console output and full history. Follow-ups resume the same session. - 👀 Real PR reviews: Catalog reviewers (code, security, release and more) work in their own worktrees and post real review comments. Address-reviewers rolls findings into a follow-up. - ⚡ Triggers and screening: GitHub webhook events auto-start tasks with dedup, delivery log and replay. Scheduled screenings audit repos and notify only, they never auto-act. - 🔒 Private by design: Localhost-only, single owner. GitHub PAT in a 0600 vault with ingestion-level masking, gh CLI hard-blocked for agents. --- OPEN SOURCE AND FREE TO TRY ⚡ Jalebi is 100% open source under AGPLv3 with unlimited tasks, repos and runs. No seats, no tiers, no new subscription, it rides the CLI subscriptions you already own. Grab it at https://github.com/samosa-ai-com..., run ./start.sh, open http://127.0.0.1:2052, add your GitHub PAT, connect a repo, and file your first task. Linux-first (Ubuntu 22.04+, Debian 12+, Arch, Fedora). --- WE'D LOVE YOUR FEEDBACK! 💬 Running LLM agents against real repos comes with sharp edges across CLIs, models and repo layouts. We'd love to hear from you: 1. Which CLI backend and model pairing are you running with Jalebi? 2. What GitHub workflows would you automate first (issue fix, PR review, nightly audit)? 3. Any worktree, webhook-tunnel or merge-gate quirks you spot on your setup! GitHub: https://github.com/samosa-ai-com... Docs: https://samosa-ai.com/jalebi/docs Landing: https://samosa-ai.com/jalebi My co-founder and I will be hanging out here all day to answer questions, chat about localhost-first agent tooling, and take feature requests. Thank you for checking out Jalebi! ❤️

Jalebi

Your repos. Your agents. Your rules.

Your repos. Your agents. Your rules.

What's great

vs Alternatives

What's great

vs Alternatives