Show HN: Orchestra – an interface to usefully run coding agents in parallel (github.com)

Orchestra is a new TUI for running coding agents in parallel: you describe a feature, a “designer” agent decomposes it into tasks and spawns isolated “executor” agents (each in a container by default) to implement them concurrently. Executors run via Claude Code, have terminals and diffs you can inspect, and a background monitor enforces the spec so agents don’t drift. You can jump into any executor, view and stage its changes locally (pairing mode), edit the shared spec, and merge the best results — enabling workflows like best-of-n implementations, simultaneous independent feature work, and iterative plan decomposition. Its significance is practical: Orchestra targets the core challenge of scaling agent-driven development safely by combining model-to-model orchestration, monitoring, and a human-in-the-loop merger/oversight step. Technically it’s a three-pane tmux/textual UI with Docker-backed executor isolation (configurable in ~/.orchestra/config), integrates with git/claude-code/python/docker, and is installed via pip (pip install orchestra-code) with a setup flow (orchestra-setup) and run command (orchestra). Shortcuts and commands (spec editor, /merge-child, pairing toggle) speed interaction. Known gotchas include MCP server/port, Docker/auth issues and message delivery variability; the project asks for bug reports and contributions. Overall, Orchestra aims to make parallel coding agents productive and auditable rather than chaotic.