ADE

What it does

• Organizes your work into teams, each holding named agents with their own photos.
• Gives every agent its own git repo or worktree, so agents don’t trip over each other’s code.
• Opens each agent in real terminal tabs running its coding CLI, with as many parallel sessions as you want.
• Lets you start a session on a different model from a row of logos: Claude, OpenAI, Kimi K2.7, MiniMax M3 or GLM 5.2.
• Keeps a memory for each agent in plain markdown files (who it is, who you are, its notes and its skills) that grows as it works.
• Shows that memory in an Agent Files panel so you can watch it fill in and open any file.

How it works

1. Create a team, then add an agent with a name, an optional role, a runtime (Claude, Codex or OpenCode) and a repo.
2. Open the agent and a terminal session starts automatically in its own worktree.
3. Click a logo in the model bar to open another session on a different model, same code.
4. As you work, the agent writes what it learns to its memory files, and at the end of a session it reviews the conversation and updates them, so the next session starts smarter.

What you need

• A Mac
• Git (install Apple’s command line tools with xcode-select --install)
• At least one coding CLI. Claude Code is recommended, because it also runs the Kimi, MiniMax and GLM sessions.
• Node.js, only to install those CLIs with npm
• An OpenRouter API key, only for the Kimi, MiniMax and GLM sessions
• Bun 1.0 or newer, only if you build from source

Get the code

Download it, or clone it and follow the steps below. You can also hand the repo to Claude Code or Codex and ask it to set it up for you.

npm i -g @anthropic-ai/claude-code
npm i -g @openai/codex
npm i -g opencode-ai

1. Download the signed DMG from the latest release on GitHub, open it, and drag ADE into your Applications folder.
2. Install at least one coding CLI. Claude Code is the recommended one. Codex and OpenCode are optional.
3. Launch ADE and click Create a team, then hover the team in the left rail and click + to add your first agent.
4. The first time you pick Kimi, MiniMax or GLM, paste your OpenRouter key and choose Save & Launch. You only do this once.
5. To build from source instead: git clone the repo, cd into damon-ade, run bun install, then in apps/desktop run bun run compile:app and bunx electron .

The full setup guide is in the README on GitHub.

Try it

• Make a team called Newsletter with an agent called Scout, leave the role blank, and shape it by talking to it.
• Open the same agent in a second session on GLM 5.2 from the model bar and compare it with the Claude session on the same code.
• Tell the agent a preference or correct it, then watch its notes update in the Agent Files panel.

Security

• Nothing is hosted. Your code, keys and agent memory all stay on your Mac.
• Your OpenRouter key is encrypted with the macOS keychain, stored locally, injected only into the agent’s terminal, and never shown back in the app.
• Claude and OpenAI sessions sign in through their own CLIs, so ADE doesn’t hold those keys.
• Agent memory files live outside your git worktree, so they’re never committed with your code.

Good to know

• It’s Mac only.
• It doesn’t include any coding CLI. You install Claude Code, Codex or OpenCode yourself.
• The Kimi, MiniMax and GLM sessions need Claude Code installed plus an OpenRouter key.
• The end-of-session memory review is enforced on Claude Code. On the other runtimes it runs by convention.
• When building from source, the README warns against electron-vite preview because it can run out of memory.
• It’s a modified version of Superset and is licensed under the Elastic License 2.0.