Meevra Code Download

Free Public Preview · Windows

AI coding agents
you can check.

Meevra Code is a Windows app that runs AI coding agents on your own projects, with the models you connect. They ask before they change anything, and Meevra checks their work before you trust it.

Download for Windows

Windows 10/11, 64-bit Free Release notes

Meevra Code runs on Windows PCs.

Send yourself the link and download it on your computer.

Release notes

The preview installer is not code-signed yet, so Windows SmartScreen may show “unknown publisher”. Choose More info, then Run anyway. Installed copies update themselves.

How it works

Your projects. Your models. Your call.

  1. Connect a provider

    Sign in to Claude Code, ChatGPT or Grok Build with the maker's own sign-in, add an API key or a coding-plan key, or run a local model. Meevra Code includes no AI model of its own, so you use the providers you already trust.

  2. Pick a folder

    Choose a folder on your computer. It becomes your project, and Meevra Brain learns it on your computer, without a model.

  3. Describe the task

    A Lead agent plans the work, hands scoped parts to helper roles and puts the result together. By default they ask before any step that changes files, runs commands or commits code.

Proof, not promises

“Done” is a claim. Meevra checks it.

An agent can say the tests pass. Meevra Code looks at what really happened in the task, and shows you where the answer and the facts disagree.

Claims checker

Every claim the agent reports in a checkable form gets a verdict from what really happened: the commands it ran, their exit codes, the files it changed. A contradicted claim is highlighted, and Show proof opens the record that decided it.

Claims
  • Verified Unit tests pass
  • Not verified Works on the staging server
  • Contradicted Lint is clean
Illustration. The verdicts are the app's own words.

Test truth

Do the agent's new tests fail without its change? Meevra checks it itself, in a separate copy of your project with the source put back as it was, and tells you which tests really check the change. One click asks the agent to fix the ones that don't.

Test truth
  • parses an empty configChecks the change
  • renders the settings pageDoesn't check the change
  • retries on timeoutCouldn't tell
Illustration. The verdicts are the app's own words.

Proof it works

Before a task may end as done, Meevra Code runs your tests and your typecheck or build itself, and shows the exact commands and their results. A task whose checks did not pass does not read Done.

Proof strength

In a temporary copy of your project Meevra Code makes one small break at a time in the changed lines, and shows which breaks your tests noticed. None of these checks uses a model, so they cost no tokens.

Inside the app

Everything an agent does, in plain sight.

Quota, as the provider reports it

Each provider's limits and balances, exactly as the provider reports them, and a quiet quota line under a running task. Meevra's own token counts are labelled as local estimates, never shown as a provider limit.

When a plan is used up, Meevra stops and asks. Nothing moves to pay-as-you-go billing on its own.

Provider limitsas reported

Meevra Brain

A project memory that lives on your computer. It reads your files, Git history and project settings on its own, without a model, and keeps up as they change.

Mission Control & Flow

Who works on what right now, what waits for you, and every project's workflow: each agent, its model and speed, the file it works on and its tokens.

Computer use

Agents can take screenshots and click, type and scroll in the apps you allow, with the model you choose. The screen edges glow while they work and you can stop them at any time. Screenshots stay local unless you confirm a cloud destination.

Approvals & Activity

You choose what agents may do. Every step of a task is listed as it happens: tools, approvals, tests and provider events.

Local models

Run models on your own machine with Ollama, LM Studio or the built-in local engine, and mix them with cloud models in one crew.

Tools & MCP

MCP servers give the crew extra tools, such as GitHub, a database or a browser. Bring in the ones you already use in other apps.

Vault

Notes and reference text for your chats and code tasks, encrypted on your computer.

A browser and a terminal beside your work

Open a terminal or a browser in a pane next to the page you are on. The browser has tabs and keeps its sign-ins on your computer. Ask for it, and an agent works in that browser in its own tab while you watch.

Built-in skills

Thirteen skills ship with the app: video prompting and editing, image generation, editing and upscaling, SVG and logo quality, website design, model comparison. An agent reads a skill only when the task needs it.

Delivery

From a finished task to a pull request: Meevra Code pushes only the files the session changed, opens the pull request on GitHub and reads its checks. It asks you first, and uses no model.

Reply

Answer any message of a thread with a quote, as in a chat app, with every model. A reply sent while a task runs reaches that task at its next step.

Stuck detector & error catcher

A task that goes in circles is paused early, with a plain explanation, before it burns your quota. Errors from the commands and pages Meevra Code starts become cards mapped to your files.

Where it went

Every model turn by model, effort, task and day, with what the API would charge for the same tokens. A running task shows how many turns tasks like it usually take, from your own record.

Works with

Sign in to Claude Code (Anthropic's own sign-in), to ChatGPT (OpenAI's own sign-in) or to Grok Build (xAI's own sign-in), and work with the account you already have there. Coding plans by their own key: Z.ai GLM Coding Plan, MiniMax M Plan, BytePlus ModelArk Coding Plan, StepFun Step Plan and the Xiaomi MiMo Token Plan. API keys for Anthropic, OpenAI and Xiaomi MiMo; OpenRouter, for models from Google, xAI, Mistral AI, DeepSeek and many more labs. Local models through Ollama, LM Studio or the built-in engine.

Meevra Code never sees your password: you sign in with the maker's own sign-in, and Claude Code ships inside Meevra Code unmodified. It resells nothing and adds nothing to what your provider charges.

Meevra Code is independent and not affiliated with or endorsed by these providers. Their names belong to their owners. Your use of a provider follows that provider's terms and prices.

Local first

Your code stays yours.

  • No account needed, and no usage analytics unless you choose to share anonymous totals (off by default).
  • Projects, chats, settings and memory are stored on your computer.
  • Your API keys stay on your computer and are never sent to us.
  • Prompts and files go straight from your computer to the provider you choose.
  • The browser pane keeps its sign-ins and site data on your computer.

The only exception is the update check. The Privacy Policy says exactly what it sends.

More in the app

It was “Coming soon”. Now it is here.

Earlier previews marked these as “Coming soon”. They are part of the app today.

  • Sign-ins

    Sign in to Claude Code, ChatGPT and Grok Build, each with its maker's own sign-in, and see the limits they report in Quota.

  • Race

    Several models do the same task, each in its own copy of the project. You see measured facts for each, and take one version or combine them file by file.

  • Night shift

    Queue a task for the night. It runs on plan quota that would otherwise be lost before the reset, never on a route that costs money.

  • Bots

    Bots that tidy, summarise or digest your folders, with your approval.

  • Mobile

    Pair your phone with a QR code and follow your tasks from it.

  • Brain Explorer

    Browse, search and edit everything Meevra remembers, and what each agent remembers about a project.

  • TV & second screen

    Extend or mirror Meevra Code to a TV or a second screen. Experimental.

What's next

Plans, not promises. No dates: each part ships when it is ready.

  • Signed installer

    A code-signed installer, so Windows no longer shows “unknown publisher”.

  • Meevra Link

    Several computers working together, so you can start a task on one and resume it on another.

  • TV leaves Experimental

    After it has been tested on real devices.

FAQ

Questions, answered.

Is it really free?

Yes. Meevra Code is free during the Public Preview, for personal and commercial use. Paid plans may come later; their price and terms will be shown before any purchase, and nothing is ever charged without your explicit purchase.

Does it include AI models?

No. You sign in to Claude Code, ChatGPT or Grok Build with the account you already have, connect providers with API keys or coding-plan keys, or run local models. Any fees a provider charges are between you and that provider.

Why does Windows warn me about an unknown publisher?

The preview installer is not code-signed yet, so Windows SmartScreen may show “Windows protected your PC”. Choose More info, then Run anyway. Every later update is verified by the app before it installs.

How do updates work?

Installed copies check for new versions on their own. When one is ready, Settings › Updates offers it: it downloads when you choose Update and installs when you choose Restart.

Is there a Mac, Linux or phone version?

Not yet. Meevra Code runs on Windows 10 and 11 (64-bit). You can follow your tasks from your phone: pair it with a QR code on the Mobile page.

Will the agents change my files without asking?

Not by default. Agents ask before any step that changes files, runs commands or commits code. You can widen or narrow what they may do in Settings, and every step is recorded in Activity.

What does “Public Preview” mean?

The app works, but it is still changing: features may change, arrive or go. Please keep backups of your projects, and tell us what breaks at [email protected].

Try it on your own project.

Free during the Public Preview. Windows 10/11, 64-bit.

Meevra Code runs on Windows PCs.