Codex and Claude Code. Same canvas, one screen.
Run OpenAI Codex and Claude Code as neighbouring terminal nodes, route each task to the agent that is best at it, and compare their work without alt-tabbing.
Two agents. Too much window switching.
- The comparison is another tab — You switch away from Codex to read Claude Code's answer. Then you switch back to check what changed.
- You carry every handoff — You ask Codex to implement and Claude Code to review. You copy the first agent's output into the second terminal yourself.
- Which session needs you? — You have four sessions open and another task in progress. You check each terminal to find the agent waiting for your input.
Give each task to the agent you choose.
Place Codex and Claude Code in neighbouring terminal nodes on an infinite, zoomable canvas. Compare their output side by side, then wire a belt to hand one agent's output to the other. SpeakCode detects turn completion for both; opt into Claude Code's Stop/Notification hooks or Codex's notify hook for exact status.
- Real terminal nodes host Codex and Claude Code.
- Live node status shows idle, working, waiting, or exited.
- The agent inbox gathers every agent waiting for input.
- Belts deliver one agent's output to another when its turn finishes.
Both agents in view. Hand the work across.
| Feature | Two terminal windows | SpeakCode |
|---|---|---|
| Both CLIs unchanged | Yes | Yes |
| See both agents' output at once | if you tile them | Yes |
| Know which agent is waiting | read each screen | live status and agent inbox |
| Hand Codex's output to Claude Code and back | copy and paste | belts deliver output when each turn ends |
| Share the session with a teammate | screen share | real state replication |
How it works
- Subscribe and install — Subscribe for $20/month or $200/year. Download the macOS app from your account page and sign in to reach the canvas in about two minutes.
- Bring both agents — Run Codex and Claude Code in their own terminal nodes with your OpenAI and Anthropic accounts or API keys. Keep your CLAUDE.md, settings, and MCP servers.
- Conduct from one canvas — Drag your agents side by side and give each a task. Zoom out for live status, zoom in to type, and connect a belt when one agent should hand output to the other.
FAQ
Does SpeakCode replace Codex or Claude Code?
No. SpeakCode hosts the real CLIs in real VT100/xterm terminal nodes. You keep your CLAUDE.md, settings, and MCP servers, and can type directly into either terminal.
Do I need both OpenAI and Anthropic subscriptions?
You bring your own OpenAI and Anthropic accounts or API keys to run their agents. SpeakCode's $20/month subscription covers the app; AI usage is billed by the providers.
Can Codex hand work to Claude Code?
Yes. A belt is a directed connection between cards. When the source agent finishes its turn, the belt transforms its output and delivers it into the target, with actions to press Enter, run a command, or paste the last N lines. You can wire Codex's implementation output into a Claude Code review.
Is there a free trial, and how do I cancel?
There is no free trial. One plan includes every feature for $20/month or $200/year, with two months free on annual billing, approximately $16.67/month. Payments run through Stripe and subscriptions auto-renew. Cancel renewal anytime from the account page through the Stripe billing portal. Access continues until the end of your paid period; prepaid time is not refunded.
Which operating systems does SpeakCode support?
SpeakCode is a native macOS app and requires macOS 26 or later. It is not available for Windows or Linux.