# LoopTroop: Local open-source GUI for long AI coding tasks (LLM council + Ralph loops)

**URL:** <https://forum.cursor.com/t/looptroop-local-open-source-gui-for-long-ai-coding-tasks-llm-council-ralph-loops/171346>\
**Category:** Built with Cursor\
**Tags:** context\
**Created:** [September 11, 2026, 12:11pm UTC](https://forum.cursor.com/t/looptroop-local-open-source-gui-for-long-ai-coding-tasks-llm-council-ralph-loops/171346 "2026-09-11T12:11:42Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![Stefanescu\_Liviu](https://sea3.discourse-cdn.com/cursor1/user_avatar/forum.cursor.com/stefanescu_liviu/32/120578_2.png) [@Stefanescu\_Liviu](https://forum.cursor.com/u/Stefanescu_Liviu)\
**Post date:** [September 11, 2026, 12:11pm UTC](https://forum.cursor.com/t/looptroop-local-open-source-gui-for-long-ai-coding-tasks-llm-council-ralph-loops/171346/1 "2026-09-11T12:11:42Z")

</div>

Hi everyone, wanting to share an open-source project I’ve been building called LoopTroop. It’s a local-first GUI app designed for running multi-step coding tickets without hitting context rot.

![20260619104032-26sec-captions-ezgif.com-resize](https://us1.discourse-cdn.com/cursor1/original/3X/7/2/727eeee870b95b9dbb3ea9bde79e74856fa7a519.gif)

Instead of letting a single model write code in one massive, noisy prompt, it takes a slow and precise approach:

- **LLM Council planning:** Multiple model passes draft, review, and anonymously vote on execution plans before touching any code.
- **Context engineering:** Each task step runs in a fresh, isolated context window with only the code it needs, avoiding bloated prompt histories.
- **Ralph loops:** If a step fails tests or checks, a clean retry loop runs with targeted failure logs instead of appending everything to a giant prompt.
- **Human in the loop:** You can review plans, steer execution, and approve steps on a modern Kanban board GUI.

Just passed 130 stars on GitHub and I’m working on it daily.

- GitHub: [GitHub - looptroop-ai/LoopTroop: Local AI coding orchestration for repo-scale work: LLM-council planning, Ralph-loop recovery, isolated OpenCode worktrees, and human-gated PR delivery. · GitHub](https://github.com/looptroop-ai/LoopTroop)
- 2m30s video demo: [https://youtu.be/g1A2g-oOR3E](https://youtu.be/g1A2g-oOR3E)

Any feedback is more than welcom 🙂

---

<div class="post-metadata">

**Author:** ![izgorodin](https://sea3.discourse-cdn.com/cursor1/user_avatar/forum.cursor.com/izgorodin/32/117016_2.png) [@izgorodin](https://forum.cursor.com/u/izgorodin)\
**Post date:** [September 12, 2026, 6:05am UTC](https://forum.cursor.com/t/looptroop-local-open-source-gui-for-long-ai-coding-tasks-llm-council-ralph-loops/171346/3 "2026-09-12T06:05:01Z")

</div>

The three separate retry-note histories make a useful test possible: let attempt one blame a missing dependency, then have attempt two verify the dependency is present and expose a different failure. On attempt three, does the old diagnosis still steer the agent? I’d preserve it in the audit trail but mark it superseded in the retry context, so a clean session doesn’t keep inheriting a disproved explanation.

---

<div class="post-metadata">

**Author:** ![Stefanescu\_Liviu](https://sea3.discourse-cdn.com/cursor1/user_avatar/forum.cursor.com/stefanescu_liviu/32/120578_2.png) [@Stefanescu\_Liviu](https://forum.cursor.com/u/Stefanescu_Liviu)\
**Post date:** [September 13, 2026, 6:49am UTC](https://forum.cursor.com/t/looptroop-local-open-source-gui-for-long-ai-coding-tasks-llm-council-ralph-loops/171346/4 "2026-09-13T06:49:53Z")

</div>

Huh?