All seven do the same basic job: read your folder, take a request, change files, run commands, ask before anything risky. The differences are the model behind them and the account they sign in with. If you have a Claude plan, start with Claude Code. A ChatGPT plan, Codex. Neither, OpenCode with whichever provider you have a key for.
| Agent | Made by | Signs in with | Good for | |
|---|---|---|---|---|
| Claude Code | Anthropic | Your Claude account (subscription) or an API key | Long, careful jobs across a whole codebase. Slash commands, skills and MCP connections all work inside the pane. | Docs |
| Codex | OpenAI | Your ChatGPT account | The natural pick if you already pay for ChatGPT. Quick edits and test-fix loops. | Docs |
| OpenCode | open source | Any model provider you have a key for | Bring your own model, including local ones. Good when you want to choose the model per job. | Docs |
| Grok | xAI | Your xAI account | xAI's coding agent. Pick it if Grok is the model you already use. | Docs |
| Antigravity | Your Google account | Google's coding agent, the successor to Gemini CLI. Uses your Google sign-in. | Docs | |
| Hermes | Nous Research | A model provider key, set up on first run | An agent that keeps notes as it works, so the next job starts with what it learned on the last one. | Docs |
| Kimi Code | Moonshot AI | Your Kimi account | Moonshot's coding agent. Pick it if Kimi is the model you already use. | Docs |
Press Ctrl+N. Agents already on your computer are marked ready. A missing one opens a setup sheet with the vendor's own install command; press Install it for me and it runs in a terminal pane in front of you. When it finishes, the launcher updates without a restart. You can also install any of them yourself, the way its docs describe, and MoveMyCode will find it at the next launch.
Nothing stops you. A common desk is Claude Code on the main job and Codex or OpenCode on something beside it, like release notes or a test run. Each pane is its own process with its own sign-in, and they share nothing except the folder.
OpenCode is open source and takes a model key from most providers, including ones that run on your own machine. If you want to try a model none of the branded agents ship with, that is the route.
Sessions are per agent, so switching means starting a new session with the other one. Your folder, your library and your connections stay the same. The connections MoveMyCode writes for one agent are copied to the others that read the standard config, so an MCP server you set up for Claude Code is there for OpenCode too.