Pane is not an AI coding agent. It doesn't write code, generate suggestions, or call LLMs. Pane is the tool you use to run, manage, and orchestrate AI coding agents like Claude Code, Aider, OpenCode, Goose, Gemini CLI, and CodexCLI — in parallel, each in its own git worktree.
| agent | what it does | works in pane? |
|---|---|---|
| claude code | anthropic's CLI agent — edits files, runs commands, builds features | yes — just run it |
| aider | AI pair programming in the terminal — works with any LLM | yes — just run it |
| opencode | go-based terminal AI agent with TUI — 112k+ GitHub stars | yes — just run it |
| goose | block's open-source agent — runs code, calls APIs, MCP-native | yes — just run it |
| gemini CLI | google's CLI agent for gemini models — 70k+ GitHub stars | yes — just run it |
| codex CLI | openai's rust-based terminal coding agent | yes — just run it |
| any future CLI agent | whatever ships tomorrow | yes — just run it |
agents write code. pane manages agents.
Think of Pane like a cockpit and agents like pilots. The cockpit doesn't fly the plane — it gives you visibility and control over everything that does. Pane runs agents in isolated terminals, shows you their diffs, and lets you commit/push/merge from the keyboard.
one agent is fine. five agents need pane.
If you only run one agent on one task, you can use the agent directly in any terminal. But the moment you want to run multiple agents in parallel — Claude Code on feature A, Aider on feature B, Codex on bug fix C — you need something to manage them. That's Pane.
pane doesn't wrap agents
Other tools build custom chat UIs around agents, which means they need to add explicit support for each one. Pane gives every agent a real terminal. If it has a CLI, it works — instantly, with zero integration, no SDK, no plugin. The agent doesn't even know it's running inside Pane.