Ordewell turns one goal into an ordered plan of coding-agent tasks, each with its own runner, model and mode, then executes and verifies the results. The plan is a typed artifact, not an agent's internal state, allowing changes before tokens are spent without losing completed work. The planner assigns the right model per task and shows every assignment before anything runs. A task completes only when its unique completion marker appears in the runner's output; exit codes are retained as diagnostic evidence. Planning is one continuous chat that researches your repo read-only, asks when goals are vague, and its final message is the plan. No extra API key required — Claude Code, Codex, or Open Code can be the planner, strictly read-only, on the subscription already held for the runners. Multi-runner by design, supporting Aider and custom CLIs via plugin manifests. Quick start requires Node.js ≥ 20 on macOS, Linux, or Windows, with tmux for the TUI. Installation is via `npm install -g ordewell` or the VS Code extension. Scriptable and headless modes skip the TUI using envvars. Twenty-five providers are recognized via their own API keys, including OpenRouter, Anthropic, and OpenAI. Three surfaces — VS Code, Terminal UI, and headless — share one core.

The VS Code extension provides a streaming timeline with live thinking, research steps, and expandable task cards. Tasks can be retargeted with new runners, models, and modes that re-derive in place. The terminal UI mirrors the extension over SSH with tab-based pane swapping and single-key plan controls. Headless mode runs via subcommands with environment variables for planner and runner configuration.

Mutation always stays with the runners; the planner agent only explores and reasons. Built-in support includes Claude Code, Codex, and Open Code, while other tools integrate as plugin manifests without code changes.