Run ten agents at once. Lose track of none.
Every Claude Code or Codex session gets its own terminal node on one zoomable canvas. Live status on each, an inbox of the ones waiting on you, and belts that hand work from one agent to the next.
More sessions compete for your attention.
- Your status check is tab-juggling — You cycle through terminals to see which agents are working, waiting or done. Each new session adds another place to check.
- One yes holds up the work — You focus on one agent while another waits twenty minutes for your permission. You only notice when you switch back to its terminal.
- Every hand-off goes through you — You copy one agent’s output into the next session, then check which task can start. You spend your attention moving work between agents.
See who needs you. Connect what follows.
You give each Claude Code or Codex session its own terminal node. Zoom out to see the fleet, then zoom in and type directly into the session that needs your attention.
- Live node status shows idle, working, waiting or exited.
- Agent inbox gathers waiting agents with a standup-style summary.
- Belts transform and deliver output when an agent finishes its turn.
- Task cards form a dependency graph to identify ready tasks.
See the fleet. Connect the next turn.
| Feature | tmux or terminal tabs | SpeakCode |
|---|---|---|
| Many sessions visible at once | if you tile them | Yes |
| Status per session | read each screen | idle, working, waiting, exited |
| Inbox of sessions waiting on you | No | Yes |
| Hand-offs between agents | copy and paste | when an agent finishes its turn |
| Task dependency graph | No | Yes |
| Notes, browsers and editors beside sessions | in separate panes or windows | cards on the same canvas |
| Claude Code and Codex side by side | Yes | Yes |
How it works
- Subscribe and install on your Mac — Subscribe for $20/month or $200/year, with every feature included. Download from your account page and sign in; you can be on the canvas in about two minutes.
- Bring the agents you already use — Run Claude Code and Codex in separate terminal nodes, using your own provider accounts or API keys. Choose each agent’s directory or worktree yourself; your providers bill AI usage separately.
- Conduct the next turn — Arrange your nodes and check the inbox for agents waiting on you. Wire belts for hand-offs and connect task cards into a dependency graph so SpeakCode can decide what is ready to run next.
FAQ
How does SpeakCode know an agent is waiting?
You get detection of turn completion, input requests and exits without configuration. Each node shows idle, working, waiting or exited, plus an Activity feed. The agent inbox lists agents waiting for input across all nodes with a standup-style summary. Opt into Claude Code’s Stop/Notification hooks or Codex’s notify hook to make status exact.
Can agents hand work to each other?
Yes. You draw a belt, a directed connection between two cards. When the source agent finishes its turn, its output is transformed and delivered into the target. You can configure delivery to press Enter, run a command or paste the last N lines. You can also connect task cards into a dependency graph SpeakCode uses to decide what is ready to run next.
Does SpeakCode manage git worktrees for me?
No. You run each agent in whatever directory or worktree you choose. SpeakCode hosts the real terminals; you create and manage your worktrees yourself.
Is there a free trial, and can I cancel?
There is no free trial. You pay $20/month or $200/year for one plan with every feature included. Your subscription auto-renews. You can cancel renewal anytime from the account page through the Stripe billing portal. Your access continues to the end of the paid period; prepaid time is not refunded.
Can I run SpeakCode on Windows or Linux?
No. You need macOS 26 or later. SpeakCode is a native macOS app and is not available for Windows or Linux.
Further reading
- Belts: hand one agent’s output to the next automatically
- Git worktrees for parallel coding agents
- How to run multiple Claude Code sessions
- What Claude Code Actually Costs When You Run Agents in Parallel
- Claude Code Agent Teams: Running Parallel Agents That Don't Collide
- Claude Code Workflows That Actually Ship Code (vs. Chat With Extra Steps)