Mework documentation

Mework is a desktop coding agent for macOS and Windows that lets you steer what the model sees and does: the context, the tools, and every prompt Mework adds. You bring your own provider key or subscription; there is no Mework account, and nothing leaves your computer except the requests listed in the privacy policy.

Install

The links point to dl.mework.dev, Mework's download mirror. The same files are on the GitHub release page.

macOS

Open the .dmg and drag Mework to Applications. The app is signed with an Apple Developer ID and notarized by Apple.

The first time Mework needs its keychain item, for example when you save an API key, macOS asks whether it may use Mework Safe Storage. Choose Always Allow so macOS doesn't ask again for this version of Mework.

Windows installer

Run Mework_1.1.0_x64-setup.exe. It installs Mework for all users, so Windows asks for administrator approval. If WebView2 is missing, the installer downloads it, which needs an internet connection.

Windows portable

Unzip Mework_1.1.0_x64_portable.zip into a folder of its own and run mework.exe. Keep every file from the archive together: no model request works without mework-aisdk.exe, the Claude Agent provider needs claude.exe, and SSH machines, WSL and the sandbox need the remote-agents folder.

Portable means no installer, not no data: the portable app writes to the same data folders and Windows Credential Manager entries as the installed one. WebView2 must already be installed.

MSIX package

The .msix is signed with Mework's own certificate, so Windows accepts it only after you trust that certificate once:

1. Download Mework_msix_signing.cer from the table above.

2. Import it into Local Machine → Trusted People: double-click the file, choose Install Certificate…, pick Local Machine, then Place all certificates in the following store and Trusted People. Or, in an administrator PowerShell:

   Import-Certificate -FilePath .\Mework_msix_signing.cer -CertStoreLocation Cert:\LocalMachine\TrustedPeople

3. Open Mework_1.1.0_x64.msix and install it.

The MSIX edition never downloads updates itself. Installed from the release page, its Updates page points you back there for the next .msix; installed from the Microsoft Store, it is kept up to date by the Store.

Note The installer, portable and MSIX editions share one data folder, so only one of them can run at a time.

Verify a download

SHA256SUMS lists the SHA-256 hash of every file in the release. Download it next to your file.

On macOS:

cd ~/Downloads
shasum -a 256 -c SHA256SUMS --ignore-missing

Each file you have prints OK.

On Windows, in PowerShell:

Get-FileHash .\Mework_1.1.0_x64-setup.exe -Algorithm SHA256
Select-String Mework_1.1.0_x64-setup.exe .\SHA256SUMS

The two hashes must match; Get-FileHash prints them in upper case.

First steps

A fresh install opens on a new task with no project chosen. Two providers are built in: Claude Agent, which runs the Claude Code that ships with Mework and is turned on, and OpenAI Codex, which stays off until you sign in with ChatGPT.

1. Get a model. If you already use Claude Code and are signed in, Mework notices on first launch, because both share one login: it lists the newest Fable, Opus, Sonnet and Haiku models under Claude Agent and selects Opus, so you can skip to step 3. Otherwise open Settings → Providers → Model providers and do one of these:

   • Claude Agent: choose Open a terminal to sign in, finish the sign-in there, then choose Re-check. You don't need Claude Code installed yourself.
   • OpenAI Codex: choose Sign in with ChatGPT and approve in the browser. Then choose Fetch models and click the models you want to add; Codex has none until you do.
   • Any other service: choose Add provider, give it a name, pick its Chat protocol and choose Add. Paste the key into API Key (it is saved when the field loses focus), check the API address, then Fetch models and click the models to add.

   Models and providers covers every protocol and model setting.

2. Pick the model in the model menu under the message box. The choice applies to every conversation's next request.

3. Add a project. In the sidebar, choose New project beside Projects, click Choose a directory… to pick your folder, and choose Create project. Mework opens a new task in it. If you skip this, the task goes to the Temporary project, which gives it an empty scratch folder that is deleted with the task. See Workspaces and machines.

4. Check the security level. New tasks start at Manual: the model reads files in your project without asking, and every file write, shell command and web search waits until you approve it on an approval card. See security levels.

5. Send. Type your request and press Enter; ShiftEnter starts a new line. You can swap these in App settings.

The window

The sidebar is on the left and the timeline in the middle, with the message box below it; the security level, Plan, model, reasoning effort and context ring sit under the message box. Side panes open beside the timeline from the buttons in the top bar and its More options menu.