Radiant

100% free and open source. MIT licensed — use it, modify it,
  redistribute it, sell it.

  A local coding harness for your Mac: chat with coding agents across cloud and
  local models, watch them work in a live activity feed, and drive a real
  terminal — all in one window.

  An agent working in a real folder — every file it touched is listed, and the
  composer shows which model is answering and what it is allowed to do.

  Longer jobs run as tasks. "Needs you" is the column that matters: an agent
  that hits a decision stops there instead of guessing.

  A loop is a run of steps that checks its own work. Each step carries a
  condition — a shell command that has to exit 0, or a sentence another agent
  judges — and a step that fails goes round again carrying the reason.

  A graph is several jobs and only the waits that are real. You tick what a
  step actually reads; everything left unticked runs at the same time. Here that
  is four stages, three of them running together.

  Local models are measured against the machine you are on, not a spec sheet
  — every one is labelled Runs well, Runs tight or Won't run before you spend the
  download.

Features

Agents that do real work

- Agent chat with streaming replies and visible model thinking, on any
  model: Anthropic, OpenAI, OpenRouter, xAI, Nous, Groq, Mistral and more, or
  Ollama and LM Studio on your own Mac. Sessions store messages in a neutral
  format, so you can switch models mid-conversation and keep your context.
- Real tools — read, write and edit files and run shell commands in the
  chat's project folder. Ask before every command, auto-run the low-risk ones
  (with a second opinion on each from a small decision model), or allow all.
- Undo — every reply that edits files leaves a checkpoint; roll the…