Verbative
3 min read

Run several Claude Code agents in parallel in VS Code — and conduct them by voice

Run multiple Claude Code CLI sessions side by side in VS Code, switch between them by voice, and keep them aligned with one shared on-device memory.

Pairing with a single AI is already old news. The way you actually ship now is with a team of agents — one refactoring the API, one writing tests, one chasing a bug — all at once. The problem is that running several agent sessions is a tab-juggling, keyboard-bound mess, and your attention is the bottleneck.

Verbative makes your voice the conductor. Instead of living inside one chat window, you run the whole room.

Talk to any agent, hands-free — just say the word

Group your Claude Code CLI sessions into channels — a channel per project, feature, or area — and run several agents in parallel inside each one. Whatever you say goes to whichever agent has focus, so switching who you're talking to is a single click or a spoken command — not a hunt through windows.

The multiplier: every channel shares one on-device memory. A decision one agent makes is recalled by all the others — so the team never re-explains context to itself, and you never re-explain it to the team.

Verbative
Code Advanced · unlimitedAccount ↗
Signed in as mia@studio.dev
Sign out
UsageWeek 44%Fable 88%
📁 project
Web App · Checkout flow
Idle — waiting for voice command
Speed 1.3×
Voice queue
Web App, Checkout flowBuilt the checkout payment flow; all 18 checkout tests pass.Last▶
Channels
Project memory
Web App✎✕
Checkout flow✎▷↺✕
high▾
Onboarding✎▷↺✕
high▾
+ Add an agent
Mobile App✎✕
Push notifications✎▷↺✕
high▾
+ Add an agent
+ New channel

Each channel runs its own Claude agents in parallel. Your voice goes to the selected agent.

Verbative MemoryON
Verbative Mechanics
Verbative Shot
Drag a region → clipboard · Esc to finish
^⌘S — click to change
Verbative Advisory BoardOFF
ProblemsOutputDebug ConsoleTerminalPortsVerbative
Web App2 agents · 2 live · 1 working
Checkout flow
build the checkout payment flow

Wired up the Stripe PaymentIntent with an idempotency key and the charge path.

recallpayment conventions for this repo0.3s›
Editsrc/checkout/intent.ts0.6s›
Bashnpm test -- checkout2.1s›

All 18 tests pass. Edited src/checkout/intent.ts (+24 −3).

Built the checkout payment flow; all 18 checkout tests pass.
1m 12s · 3 turns
Message Checkout flow…
Idleauto mode23% ctx
Onboarding
finish the onboarding welcome step
recallonboarding copy conventions0.2s›

Built the welcome step and added form validation.

Editsrc/onboarding/Welcome.tsx0.8s›
Bashnpm test -- onboarding1.8s›

All 12 onboarding tests pass.

Built the welcome step with form validation; all 12 onboarding tests pass.
Thinking… · 611 tokens
Message Onboarding…
Working…auto mode31% ctx

One window: the Verbative panel on the left, each agent as a native chat on the right — streamed replies, tool cards, and permission prompts you answer with a click or your voice.

Several agents in one channel — your voice routes to whichever one has focus.

The vibe-coding toolkit

Around the agents sits everything that keeps you in flow without a keyboard:

  • Mechanics — turn the commands you run a hundred times a day (spin up the dev server, kick off a build, open a browser) into one-click buttons, each running in its own real terminal.
  • Verbative Shot — one shortcut drags any region of your screen straight to your agent; keep dragging to stack several into a single paste. A design to build, a render that looks wrong, an architecture sketch to scaffold from.
  • Spoken status, spoken control — say "update" for a status check while an agent works, and "stop claude" to abort a run. Replies and summaries are read back to you.
  • ~99 languages, on-device — dictate in your native language; it's transcribed and translated to English locally with whisper.cpp before it ever reaches an agent.
  • A review board when you want one — optionally send a design idea to a customizable board of domain advisors that weigh in, in parallel, before you build. Off unless you turn it on.

All local, all yours

Wake-word detection and dictation run on your own machine; the voices are local too. It runs on your existing Claude Pro or Max subscription through the Claude Code CLI — no API key, no SDK. Your audio never leaves your machine.

It adds up to a command center for vibe-coding: a whole team of agents, conducted by your voice.

See how to command a team →

Frequently asked

Can I run several Claude Code sessions in parallel in VS Code?

Yes. Verbative's Agent Workspace runs each Claude Code CLI session in its own terminal tile inside VS Code and groups them into channels, one per project, feature or area, with several agents per channel. The pricing page lists how many channels and agents each plan includes.

How do I switch between agents by voice?

Say the agent you want, by name or by its number, and whatever you say next goes to that agent. Each agent has its own status bar that shows when it is working, and spoken replies are attributed to the agent that answered when more than one is active.

Do the agents share a memory?

Yes, with Verbative Memory, which is part of Code Advanced and comes with a 14-day trial. Every channel shares one on-device memory, so a decision one agent records is recalled by the others, and the memory lives as readable files in your repository.

Does the Agent Workspace work on Windows?

Yes. Channels, the per-agent terminals and their status bars run on Windows 10 and 11 as well as macOS, inside VS Code.

On your machine, not the cloud.

Verbative is voice-first, on-device tooling for AI coding agents. Nothing uploaded.