Show HN: Kaleidoscope – A Parallel AI Agent TUI (github.com)

Kaleidoscope is a new command-line TUI that runs multiple AI code assistants in parallel, comparing their outputs in isolated git worktrees so you can pick the best result and merge it back into your branch. Launched as a tmux-based workflow for macOS, it spins up a pane per model (Claude, GPT, GitHub Copilot, etc.) using the opencode CLI, runs the same prompt concurrently, and creates feature branches/worktrees like ../<repo>-<branch>-<task>-<model> for safe experimentation. The TUI gathers branch-name, task-name, prompt, provider, and a multi-select of models; you can send targeted follow-ups with @model, merge a winner with /next (commits, --no-ff merge, push, and cleanup), or abort with /bail. Preferences and per-model usage stats persist in a .kaleidoscope JSON config, and the tool supports command autocomplete. For AI/ML and dev teams this streamlines comparative model-driven coding: faster side-by-side iteration, reproducible test runs (you run kaleidoscope with --run "npm test" or "go test"), and automated branch management that reduces manual bookkeeping. Key technical implications include safer isolation via git worktrees, deterministic execution via opencode run -m <provider>/<model> <prompt>, and easy reproduce-and-merge flows — but it also raises review and security considerations since models can commit code automatically. Currently macOS-only and requires tmux plus opencode; careful review and CI remain essential when accepting model-generated changes.