Every agent. One canvas.
CodeGrid launches the same coding-agent CLIs you already use — inside real PTYs, with no wrappers and no lock-in. Run them side by side and pick the best model for each task.
Six coding agents, plus your shell
Each pane runs the real CLI in its own PTY, connected to its own project directory.
Claude — Anthropic
Claude Code (claude).
Codex — OpenAI
Codex (codex).
Gemini — Google
Gemini (gemini).
Grok — xAI
Grok (grok).
Cursor — Cursor
Cursor's agent (cursor-agent).
Venice — aider · all models
Venice runs through aider. On first launch CodeGrid installs aider for you and asks for your Venice API key, which it saves locally with restricted permissions.
Shell — your terminal
Any shell on your PATH, for the commands you'd run by hand.
No migration. No wrappers.
Your existing setup, as-is
CodeGrid doesn't replace your tools — it launches the same workflows you already use in real PTYs.
Mix agents per workspace
Claude on the API layer, Codex on the frontend, Gemini reviewing the tests — all on one canvas, all in parallel.
Resume where you left off
Sessions can continue your most recent conversation, or open a picker to resume a specific one, when you reopen the app.
They can collaborate
The Agent Bus lets one agent message and read another's pane — see the Agent Bus →
Use the best model for every job.
Free, open source, and local-first — bring the agents you already have.
macOS · Apple Silicon · Free & open source