Runs Claude Code, Codex, Gemini, Copilot and 7 more side by side, each in its own git worktree. Review PRs, debug CI, and catch bugs at commit time. No backend, no token markup: your machine, your keys, your provider's price. Free for individual devs.
Hey Product Hunt 👋
I'm Steph. I'm an ex-GitHub, ex-Microsoft App Center and ex-Twitch engineer, and I built Klaussy because my own workday had turned into a mess: a row of terminal tabs each running a different agent, hand-managed git worktrees, and a browser full of PR tabs. Every time I switched agents, the context went with it.
Klaussy puts that whole loop in one desktop window.
Mix any agents. Claude Code, Codex, Gemini, Copilot, Cursor, Cline, opencode, Kimi Code, Ollama. Set a global default, override per terminal, or run two agents side by side on the same session.
Multi-repo sessions. Pick every repo a change touches. Same branch across all of them, one worktree each, resumable later with the conversations intact.
Repo-aware agents. On session start it maps your conventions and import graph (fan-in/out, cycles, endpoint chains) and gives agents that to work from instead of generic advice. It refreshes as the repo changes.
The Full Dev Loop. Nine phases from ticket to green PR: Plan & Discovery, Implementation, Local Review & Humanize, QA & Evidence, Create PR, Re-review Remote PR, Fix CI Failures, Resolve Review Comments, Notify when Green. Works against GitHub, GitLab, and Bitbucket.
Mirroring to Discord and Slack. A session streams into a thread. Reply from your phone to answer a prompt, approve a plan, or redirect the work, then come back to your desk and continue the same conversation on the same worktree.
Guardrails. Agents can run sandboxed in isolated Docker containers via Nemesis8. Pre-commit hooks run local agent checks over staged diffs for correctness bugs, silent failures, leaked secrets, and debug leftovers. Tab-autocomplete runs on-device through Ollama at around 100ms.
No account, no backend, no marked-up tokens. Your code goes to the agent provider you already pay for, through the CLI you already installed. We don't collect your repos, prompts, or history, so there's nothing here to sell or train on. Point it at Ollama and it works with no AI subscription at all
What I'd love feedback on: which agent combinations you'd run side by side, and whether those nine phases match how your team really ships. I'll be in the comments all day