Back to the overview

A closer look at

The real screens. The workflows. The tools around your agents.

The real app, with example projects.

Four Grid panes: Codex and Claude in terminal mode, and Codex and Claude in chat mode.
01 / 05Terminals and chats, side by side

Every task starts clean and can be resumed.

A fresh session per task keeps context small, which helps reduce hallucinations and token cost. Resume any task later.

Set a limit on parallel tasks for each provider, per project. Once it is reached, new tasks wait in the queue and start as soon as a slot frees up.

Context in one long session compared with one session per task In one long session, context and cost keep growing with every request. With one session per task, context starts from zero for each task, the conversation is saved when the terminal closes, and a saved conversation can be resumed later.
Illustration, not measured data.

Finished work waits for your review.

In one crowded session, tasks can get skipped. Here each task gets its own terminal and history, so every result waits for you.

Reviewing finished tasks Four task terminals finish one after another and each result moves into Ready for review. The first result is opened, showing its summary, changed files and passing tests, and a follow-up is sent in the same conversation, so that task runs again. The other results are marked reviewed one by one. The first task returns as turn 2, is marked reviewed, and the list ends with all results reviewed.
Sample Billing API tasks.

Two agents check the plan before code exists.

One agent drafts, the other challenges, the author revises.

Starts with

Plan request: Add invoice CSV export

  1. Claude writes the plan
  2. Codex checks it
  3. Back to Claude
  4. You hand it off
Plan council room Two agents share a desk in front of a wall display. The first writes the plan and passes the sheet to the second, who checks it and marks three findings. The sheet goes back to the first, who fixes them in version two. A path is drawn to the execution desk, which waits for your handoff.
Sample plan.

The tools around the queue.

Pick the tool you want to explore.

Your queue, in your pocket.

Pair with a QR code, then queue, watch and review from anywhere while your computer does the work.

  1. Scan the QR code
  2. Queue from your phone
  3. Your computer runs it
  4. Review from anywhere
Mobile control: pair a phone, queue a task and review it The computer shows a QR code and the phone scans it. The phone asks to connect and the computer enables access, so the computer connects out to vibeide.dev. On the phone a task is typed and added to the queue; it reaches the computer, where an agent sits down at a free desk in the Office and works on it while the phone shows the same Office. When the task finishes, the phone shows a notification without the prompt or the result. Opening it shows the task with the screenshot it captured, and the task is marked reviewed.
Optional, for projects you pick. Outbound HTTPS, no open port. Prompts, result excerpts, screenshots and project paths pass through the hosted service; notifications never include prompts or results. Needs VibeiDE running on an awake computer.

Hold. Speak. Release.

Hold your shortcut or microphone to dictate a prompt or follow-up. Release, check the text, send.

Your paired phone PWA works as a microphone and remote controller too. The speech model runs on your computer.

Optional. 670 MB download. About 800 MB RAM. No GPU.

Your voice becomes a follow-up on your computer An animated example: hold the microphone on a paired phone or computer, speak, release and review the locally transcribed text before sending. VibeiDE Follow-upSame conversation Listening... Add a search bar to the header.Ready for you to review Local speech modelRunning on your CPU CtrlShiftSpace Send VibeiDE MobileConnected to your computer Hold to talk
Desktop audio stays local. Phone audio is relayed to your computer through Mobile control. Nothing sends automatically.

Smart plans. Faster execution.

Let a capable model plan, then hand the work to a lighter model. Choose how many planners and executors run in each project.

Requests queue when planners are busy. Executors pick up plans as soon as they are ready. Execute has its own limits.

Independent planning and execution lanes Two capable planners prepare queued requests. Three lighter executors pick up ready plans. Direct Execute work runs alongside them with separate limits. Turbo modeProject limits: 2 planners / 3 executors Your requests Search the appExport invoicesFix sign-in Waiting for a planner Capable planners Planner 1Planner 2 Build the planBuild the plan Ready plans Search planExport plan Pick up immediately Lighter executors Executor 1Executor 2Executor 3 Execute Direct tasks keep moving alongside Turbo.Their own provider limits. Their own queue capacity.
Example: two planners and three executors. Every task gets a fresh planning session and a fresh execution session.

Turn finished work into a changelog.

Standup sorts finished tasks by category, ready to copy.

Completed tasks

  1. #104 PDF statement download
  2. #105 Invoice CSV export
  3. #98 Simplify date filters
  4. #106 Fix export rounding
  5. #107 Secure download access

billing-api Sep 4 to Sep 5, 2026

Added

Monthly PDF statement download #104

Invoice CSV export #105

Changed

Removed the legacy date-filter toggle #98

Fixed

Rounding errors in CSV totals #106

Security

Restricted statement downloads to account owners #107

Keep track of your provider usage.

Provider-reported 5 hour and weekly limits, with reset times.

Sample windows.

Claude, 5 hour window

42% used, resets in 2h 14m

Claude, weekly

61% used, resets in 3d 6h

Codex, 5 hour window

27% used, resets in 4h 02m

Codex, weekly

48% used, resets in 5d 12h

Keep parallel edits apart.

Two Git modes, chosen per project.

Make your work fun again.

Start on milk crates. Finish tasks, earn XP and build your dream office. Unlock 59 rewards, from comfy chairs to hologram screens and a money tree. Every level brings an upgrade. Your agents celebrate with a dance party, lights and confetti!

Achievements0 of 9 unlocked

  • Coffee’s onStart the first task of a local day. Locked
  • Quick drawReview ten tasks within five minutes of finishing. Locked
  • DiplomatUse Plan council on ten tasks. Locked
  • JugglerRun tasks in three projects at once. Locked
  • Parallel universes IdeaRun three tasks in separate worktrees at once. Locked
  • ArchaeologistResume a task older than thirty days. Locked
  • Token miserEnd a day below ten million tokens with twenty reviews. Locked
  • Well rested IdeaTake a weekend off after a five day review streak. Locked
  • Rainbow IdeaRun tasks in seven projects at once, each in its own color. Locked
From milk crates to a level 50 dream office Reviews, resumed tasks and plan councils earn XP. A door on milk crates, a crate seat and a beige CRT become a crystal hover desk, a floating throne and a five-screen hologram dome. Furniture upgrades at every level from 2 to 50. The seedling grows into a leafy plant, monstera, bonsai and money tree. A rug, neon sign, string lights, jukebox, crowns and champion trophy arrive at their real unlock levels. Agents cheer with confetti as the room levels up.
Billing API Level 1 0 / 100 XP ×1 0

0 of 59 rewardsNext: Sawhorse table at level 2

A sample year, sped up. XP comes from task metadata only, never prompts, code or transcripts. Three of the nine achievements are proposals.

Your team.
One company view.

Every teammate has a room for their agents. See shared task names, activity and work waiting for review, on desktop or phone.

Members choose what to share. Prompts, code and agent output stay private.

Explore Company mode
VibeiDE Company mode, with example teammates and their agent desks
The desktop Company view, with example people and tasks.

FAQ

Why not paste everything into one Claude Code session?

Unrelated work piles up in one context, raising the risk of mix-ups and hallucinations, and every message costs more. VibeiDE gives each task its own clean, saved session.

Why not just use tmux or separate terminals?

They work if you track everything yourself. VibeiDE adds project queues, provider limits, saved history and review, so no result gets lost in a tab.

How many tasks can run at once?

You choose. Each project has its own limit of parallel tasks for every provider, such as Codex, Claude Code and OpenCode. When a limit is reached, new tasks wait in the queue and start as soon as a slot frees up.

Do I need new API keys?

No. VibeiDE runs the Codex and Claude Code CLIs already on your computer. Your logins, keys and credits stay with them.

Does VibeiDE include Claude Code or Codex usage?

No. VibeiDE runs the Codex, Claude Code or OpenCode you already use, with your own subscription or API key. Your provider bills its own usage and applies its own limits. One provider is enough to start.

Which computers does VibeiDE run on?

macOS and Windows. The Linux build is in beta. Get them from the download page.

Can I cancel?

Yes. Cancel any time in Billing. Your plan stops renewing and access continues until the end of the paid period. The trial needs no card, so nothing renews after it.

Can I keep a terminal open between turns?

Yes. Turn on Keep open when you queue a task, then finish it when you are done.

What happens to lifetime licences if VibeiDE is discontinued?

Your licence to the last supplied version remains. Updates and hosted services may stop. The app currently requires periodic online licence validation. If development or hosted licensing permanently ends, we will notify affected customers and explain the available continuation, migration and statutory remedies. See the licence and sale terms.

Where is my data stored?

Task records are stored on your computer; prompts and context go to your AI provider. Optional Mobile control relays prompts, voice recordings, result excerpts, screenshots and project paths through the hosted service. Voice recordings are transcribed on your computer, not by a speech service. Team sync shares only the listed metadata. See Privacy.

Find your workflow

Languages

Necessary cookies support sign-in, security, your language and this choice. Optional categories stay off until accepted.

First-party page views and acquisition measurement. Google Analytics remains disabled.

Remember referral credit for later. Links still work on the current page without this cookie.

Privacy · Cookie inventory