New in 0.5.112 — out now for Windows
WieVer runs your coding agents across your projects — each isolated in its own git worktree — then merges their finished work into one integration branch that has already passed your tests.
Releases in
Download unlocked
Cancel anytime · local-first — your code never leaves your machine
$18.99/mo · runs on Claude Code, Codex, Gemini, Qwen, Grok or Kimi — your engine, your accounts
The film
The whole idea, animated: launch, fleet, isolation, the merge train, the resume — and where your one tested branch comes from. Captions included.
The problem
Running several coding agents at once is easy. Landing their work is not. Left alone they edit the same files, drift apart, and hand you a pile of branches that only conflict with each other.
The mechanism
Each session gets its own git worktree, so no two agents ever share a checkout. A foreman then walks the finished branches into one integration branch — and re-runs your test suite on the merged tree before anything is allowed to stay.
The safety net
Usage limits are account-wide, so when one session hits the ceiling every session is stuck. WieVer detects the limit, arms a resume at the reset, and fans that out to everything running — browser panes, fleet workers, and background sessions all pick up together.
Capabilities
Everything is included in the one price — there are no tiers to climb.
Dispatch a fleet of agent workers across a task and coordinate them from an overseer session that can see every worker and push the next step into any of them. Mix engines inside one run. Spend caps and a per-project allow-list bound what a run is allowed to touch, and anything destructive — a push, a branch delete, a merged pull request — waits on your explicit yes. The agent has no way to approve its own request.
Every session is launched into its own git worktree on its own branch, so concurrent agents physically cannot overwrite one another. It's the filesystem, not a convention — and it's always on, including for folders that weren't repositories when you pointed WieVer at them.
A foreman merges the finished branches in order, resolves conflicts through four escalating tiers, and gates the result on a real test run — comparing against a baseline so only new failures block the train. It also tells you whether your tests failed or the verification itself crashed: the difference between a mystery and a fix.
Sessions read as a conversation, not a console. Slash commands and file mentions from a menu, screenshots and logs dropped straight in, and hold-to-talk dictation everywhere with WieverVoice — hold Ctrl+E, say it, let go. When an agent stops to ask you something, the choices arrive as buttons you tap. Change the model or how hard it thinks mid-session, without restarting.
Twenty sessions is only workable if you can see twenty sessions. A single list shows every live one — fleet workers, the merge train, chats, workspace panes — and every row carries the same controls: switch model, change effort, open it, stop it.
Run a session on Claude Code, Codex CLI, Gemini CLI, Qwen Code, Grok or Kimi. Same worktree isolation, same merge train, whichever one you pick.
Split-pane layouts you save and reopen — terminals arranged the way you actually work, restored intact. Attach a pane read-only to watch a session with no risk of typing into it; on a phone the same layout becomes a swipeable deck.
A usage meter shows how much of your current Claude block you've spent — learned from your own transcripts and checked against Anthropic's usage endpoint, not guessed. Claude Code only: it reads Anthropic's usage API, and the other engines expose no equivalent to read.
Hit a usage limit and every running session is armed to resume at the reset — the limit is account-wide, so the whole fleet comes back together rather than one at a time. Sessions return after a restart with their conversations intact, and your sign-in is kept fresh in the background, so an unattended run doesn't quietly die overnight on an expired token.
Setup is three answers — where your projects live, a password, and how much freedom sessions get. WieVer brings its own Node and Git, kept off your system PATH so they can't collide with anything you already have, and any engine CLI you're missing installs from the Connectors screen in one click. No terminal, at any point.
Hands-on
Small things you use fifty times a day, built in so you never leave the run.
Hold Ctrl+E anywhere you could type — a composer, a dialog, a live terminal — speak, and let go. Your words land once, tidied up by AI first, and a take can never press Enter for you. One draggable pill, no mic buttons anywhere.
Ctrl+Alt+S from any app captures the screen natively; crop, annotate with pen, arrows and highlights, then copy, save — or attach it straight into the session you were briefing.
Ctrl+V an image over a raw terminal pane and it just works: the file is stored beside the session and its path is typed for you — never submitted — so the CLI reads your screenshot when you hit Enter.
Every pane sizes its own text to its own box and your device, and a two-finger pinch on any terminal — or A−/A+ in the pane menu — moves one preference that resizes them all. Background panes resize their own sessions too, so splits never tear.
Pricing
WieVer Pro
$18.99/mo
Billed in your local currency — €15.99/mo in the EU, VAT included.
Works with
The best coding CLI changes every few months. WieVer doesn't care which one you use — it launches the official CLI already installed on your machine, signed in as you, and hands every engine the same worktree isolation and the same merge train. Six engines today. When the seventh one arrives, nothing about your workflow has to change.
The same is true of your projects. WieVer drives git and a terminal — it never parses your code, builds it, or needs a plugin for your language. TypeScript, Rust, Python, Go, C#, a Unity repo, a monorepo, a decade-old codebase with a Makefile: if it's a git repository and your agent can run in it, WieVer can run a fleet on it. The only thing it needs to know about your stack is the one command that runs your tests.
Claude CodeCodex CLIGemini CLIQwen CodeGrokKimi
Whichever engine a session runs on, it gets its own git worktree on its own branch, and its finished work goes through the same merge train: merge in order, four escalating conflict tiers, then your test suite re-run on the merged tree before anything is allowed to stay. Mix engines inside one run — the train doesn't ask which CLI wrote the commit.
You use your own accounts with each provider. Nothing is resold to you, no tokens are marked up, and no prompt or line of code passes through us — WieVer runs on your machine and spawns the vendor's own CLI. Your $18.99/mo buys the orchestrator; your existing subscriptions cover the models.
default
The engine WieVer was built around — the default for new sessions, and the one the Command Deck overseer and the merge train's foreman always run on.
Models
The models your Claude plan includes, picked at launch — with an effort level.
Needs your own Claude Pro or Max plan.
Installed for you during setup if it's missing.
The only engine with a reasoning-effort dial at launch time — minimal through xhigh, chosen per session.
Models
Needs your own OpenAI account to sign in with.
WieVer installs it with its own bundled Node — nothing required on your machine first.
Four models spanning a preview flagship down to a Flash Lite for the cheap, high-volume passes.
Models
Needs your own Google account to sign in with.
WieVer installs it with its own bundled Node — nothing required on your machine first.
Three models, two of them coder-specialised — the cheap end of the roster for wide, repetitive passes over a codebase.
Models
WieVer installs it for you — no terminal.
One model, one decision. Useful when you want a second opinion on a branch rather than a whole fleet.
Models
Installed is not entitled: running it needs a SuperGrok subscription, and WieVer can't check that for you.
WieVer installs it for you — no terminal.
Four models including a one-million-token flagship — the one to reach for when a fix has to read half the repo.
Models
Wants Git for Windows present at first launch.
WieVer installs it for you — no terminal.
All product names are the property of their respective owners and are used here for identification only. WieVer is an independent product and is not affiliated with, endorsed by, or sponsored by any of them. You use your own accounts and subscriptions with each provider.
Honest about what you're buying
WieVer Pro needs your own account with whichever engine you run — a Claude Pro or Max subscription from Anthropic, or the equivalent for Codex, Gemini, Qwen, Grok or Kimi. It launches and orchestrates the official CLI you already sign in with. We never resell model access, never mark up tokens, and never sit between you and your provider's billing. Your $18.99/mo buys the orchestrator; your own accounts cover the models.
Not yet — WieVer is Windows 10/11 only for now.
Yes. The installer isn't code-signed yet, so SmartScreen warns the first time you run it. Click "More info", then "Run anyway". Code-signing is on the roadmap.
Yes — whichever engine you run. A Claude Pro or Max plan for Claude Code, or the equivalent account for Codex, Gemini, Qwen, Grok or Kimi. WieVer is the orchestrator, not a source of model access — see "Bring your own accounts" above.
Yes — Codex CLI, Gemini CLI, Qwen Code, Grok and Kimi are all launchable engines, and you can mix them inside one run. Every engine gets the same worktree isolation and goes through the same merge train. You bring your own account with each provider; see "Bring your own engine" above.
No — WieVer installs Codex CLI, Gemini CLI, Qwen Code, Grok and Kimi for you from the Connectors screen, using the Node runtime and Git it ships with, so nothing has to be on your machine first. Two honest notes: Kimi wants Git for Windows present at its first launch, and installing Grok isn't the same as being entitled to use it — that still needs your own SuperGrok subscription, which WieVer has no way to check.
Sessions render as a chat, not a console — slash commands and file mentions from a menu, attachments, dictation, and tappable buttons when an agent stops to ask you a question. A single dashboard lists every session running anywhere in the app, each row with the same stop / open / switch-model controls, so supervising six looks much like supervising one.
Yes — hold Ctrl+E anywhere you could type, speak, and let go. WieverVoice inserts what you said in one piece, optionally tidied up by AI first, and it never presses Enter for you — a dictated take can't submit a command on its own. Speech recognition runs in the app; there are no per-field mic buttons to hunt for. And for the opposite direction — showing instead of telling — the built-in snipping tool (Ctrl+Alt+S from any app) captures, crops and annotates a screenshot, then attaches it straight into a session; you can also just Ctrl+V an image into any terminal pane.
Sessions come back with their conversations intact, and you choose per session which ones restart on their own. If you hit a usage limit instead, every running session is armed to resume at the reset together rather than one at a time — and your sign-in is refreshed in the background so a long unattended run doesn't expire overnight.
No. WieVer runs entirely on your machine and binds only to localhost — there is
no relay and no cloud component. Your code talks to Anthropic directly, through
the claude CLI you already use.
Not out of the box. The desktop app is deliberately localhost-only, so remote access isn't part of the install today — it's on the roadmap. If you already run your own tunnel or VPN to that machine, the UI itself works fine in a phone browser.
Up to 20 concurrent sessions by default. Each one gets its own git worktree, so they can all work in the same repository without colliding.
The merge train merges, resolves conflicts and verifies entirely on your machine without any account. If you want it to open the pull request as well, give it a GitHub token — otherwise it leaves you a clean, tested integration branch to push yourself.
Your licence keeps working for 14 days without an internet check — enough for a flight or a bad connection. It re-verifies automatically next time you're online.
Yes — cancel from your Polar customer portal whenever you like. No lock-in.
It's emailed to you the moment your payment goes through, and it's also kept in your customer portal — sign in with the email you paid with and it's on the page. If the email hasn't shown up, check spam, then use the portal; the key is always there.
Two. Activate it on both and they work independently — a desktop and a laptop, say. If you need a third, deactivate one first; there's a button for it in Settings → License.
On the old machine, open Settings → License and click Deactivate — that frees the seat immediately. Then install on the new one and paste the same key. If the old machine is already gone or won't boot, open your customer portal and remove its activation from there; each one is labelled with its computer name.
Polar retries the charge and emails you. Nothing changes in WieVer while that's going on. If it ultimately can't be collected and the subscription lapses, WieVer stops launching new sessions — but nothing else changes; see the next answer.
They stay exactly where they are. They're ordinary git worktrees and branches in your own repositories on your own disk, and cancelling doesn't touch them. WieVer stops launching new sessions, and that's the only thing it stops: you can still open the app, read everything, run the merge train, export your work, and push. A run that's already going finishes. Losing a subscription should never mean losing access to code.
Install it on Windows, point it at your projects root, and launch your first run.
Windows 10/11 · not code-signed yet, so SmartScreen warns on first run — More info → Run anyway.
