Open Source · MIT License · v0.2.4

LLM councils plan it.
Ralph loops perfect it.
OpenCode worktrees ship it.

A smart local engine that automates big coding tasks from start to finish. Get the job done as you imagined it.

terminal
1$ git clone https://github.com/looptroop-ai/LoopTroop.git 2$ cd LoopTroop 3$ npm run dev
100% local execution Git worktree isolation Multi-model councils
localhost:5173
Demo coming soon
Why LoopTroop

LoopTroop is a local GUI orchestrator for long-running, high-correctness AI software delivery — taking you from a raw idea to merged code. Free and fully open-source.

High-speed tools optimize for immediate chat responses

They prioritize fast feedback loops — great for quick fixes, but they break down on complex, multi-file feature work where alignment and correctness are paramount. Conversation history bloats, the AI gets confused, and code quality falls off a cliff.

LoopTroop optimizes for a “slow and perfect” paradigm

It intentionally sacrifices raw speed to deliver a final result that matches exactly how you envisioned it. Instead of trusting a single, endless AI chat session, LoopTroop breaks the job into clean, separate stages.

Interview PRD Beads Ralph Loops Review
App Preview

See LoopTroop in Action

From project setup to final implementation review — explore every stage of the LoopTroop workflow. Click any screenshot to view fullscreen.

projects-dialog Click to expand
Projects Dialog — manage attached repositories, review ticket counts, and register new local git repos

Projects Dialog

Manage all your attached repositories from a central dashboard. Review ticket counts at a glance, register new local Git repos, and track the overall health of each project. The dialog gives you full visibility into which codebases are active and ready for AI-driven orchestration.

  • Register local Git repositories
  • Track ticket counts per project
  • Monitor project status at a glance
Core Primitives

The Architecture Behind Durable AI Delivery

Four foundational primitives that eliminate the common failure modes of AI-assisted coding — from biased planning to context rot and degenerate retries.

LLM Council

Multi-model planning instead of raw chat. Independent drafting, anonymous voting, and winner refinement to eliminate model bias. Each council member drafts independently, scores peers with a weighted rubric, and the winner synthesizes the strongest ideas from losing proposals.

Multi-Model Anonymous Voting Bias-Free

Beads

LoopTroop's smallest execution unit. Complex plans are decomposed into isolated, bounded coding tasks with explicit acceptance criteria. Each bead targets specific files, carries its own validation steps, and can be independently verified — preventing cascading failures across your codebase.

Atomic Tasks Acceptance Criteria Isolated

Context Engineering

Rather than passing bloated, degraded transcripts to the model, LoopTroop isolates and rebuilds minimal context from scratch for each phase. The agent only sees the specific active bead, its file target, and the test file — eliminating conversation pollution and LLM drift that plagues long coding sessions.

Minimal Context No Drift Rebuilt per Phase

The Ralph Loop

A strict manager philosophy. Instead of begging a confused model to fix its code, LoopTroop scraps degraded workspaces and retries with a fresh session inside isolated Git worktrees. Preserves a compact error trace, discards contaminated sessions, and can run 10+ hours unattended without degradation.

Fresh Retries Worktree Isolation 10h+ Unattended
Architecture

Three Architectural Layers

Each layer has a distinct core mission and technical lifecycle — separating planning from execution from shipping.

01
Planning
LLM Councils Plan It
Input Interview PRD Atomic Beads
02
Execution
Ralph Loops Perfect It
Isolated Bead Work Multi-Loop Testing & Fixing
03
Shipping
OpenCode Worktrees Ship It
Code Isolation Final Verification Main Branch
Lifecycle

How LoopTroop Works

A ticket's journey from raw idea to merged code — structured, isolated, and human-gated at every critical boundary.

01 Draft State

Define your local task ticket with a clear description of what needs to be built. This is the seed that kicks off the entire orchestration pipeline — the more precise your input, the more targeted the AI planning becomes.

Define your task ticket
02 Council Review

Multiple independent LLM models collaborate on your task. Each drafts a plan, scores peers anonymously, and votes on the best approach. The winner refines its proposal by incorporating the strongest ideas from competing drafts — eliminating single-model bias.

Multi-model voting & refinement
03 Bead Decomposition

The approved plan is split into small, logical file operations — each a self-contained "bead" with explicit purpose, acceptance criteria, dependencies, and target files. This granular decomposition prevents cascading failures and keeps each unit independently verifiable.

Atomic, verifiable tasks
04 Execution Sandbox

OpenCode executes each bead's code inside isolated Git worktrees. If execution fails, the Ralph Loop kicks in — preserving a compact error trace, discarding the contaminated workspace, and retrying with fresh context. This loop can run unattended for hours without degradation.

Ralph Loop recovery & isolation
05 Human Approval

You review every change, approve what looks right, adjust what needs tweaking, or regenerate entire beads. Only after your explicit sign-off does code get committed and merged locally. No code reaches your main branch without human verification.

Your code, your decision

Ready to Orchestrate?

Clone LoopTroop, point it at your repo, and let LLM councils plan your next feature — while you stay in control of every merge.