What it does
LoopTroop is a local orchestrator for long coding jobs. You attach a repository, create a ticket, and it moves the work through separate stages rather than one long chat. In planning, several model instances interview you, draft requirements and split them into small work units called beads, scoring and voting on each other's drafts. You approve the plan before any code is written. Execution hands each bead to the OpenCode coding agent inside an isolated Git worktree. When an attempt fails, the retry discards the session, keeps a compact error trace, and starts again with fresh context. A final verification pass and optional manual QA lead to a pull request that you review. It needs git, an authenticated GitHub CLI, and OpenCode with a configured model provider. The page strongly recommends running it inside a VM or sandboxed environment because the agent runs without permission prompts. The project calls itself early alpha software and warns that runs are slow and use many tokens by design.
Capabilities
- Multi-model planning that drafts, votes on and refines requirements
- Work split into small, test-backed units called beads
- Isolated Git worktrees keep your active branch clean
- Fresh-context retries after failed attempts
- Human approval gates before execution and delivery
- Installs through npm, Homebrew, Scoop, Chocolatey, WinGet, Docker or a standalone executable
When to use it
- You have a large multi-file feature where correctness matters more than speed.
- You want a plan reviewed and approved before an agent starts coding.
- You can leave a job running unattended, ideally on a sandboxed machine.
