Built rather than forked

The decision

Miton is its own Tauri and React desktop application, agent runtime and project system. It is not a fork of VS Code and it is not a branded layer around another agent harness.

That decision costs more than it saves in the first release. A mature editor offers deep language tooling, extension compatibility, keyboard conventions and thousands of quiet details. A mature agent harness offers a fast path to tool calling and provider support. Rebuilding those capabilities requires time and continuous parity work.

Miton accepted that cost because neither inheritance matches the product boundary.

An editor is not a neutral shell

An editor fork carries an answer to “where does work happen?” Files sit at the centre. Every other activity becomes a panel, preview or specialised file type. That is appropriate for an AI code editor. It is a structural constraint for a workspace in which a document, interface and automation are peers of code.

Miton’s answer is different. Chat is the constant place where intent begins. Canvas, Writer, Automation and Code split out when the work needs a specialised surface, then merge their results back into one project. The user may switch surfaces, but goals, knowledge, activity and cost remain attributes of the project rather than artefacts of a particular tab.

Owning the shell makes that behaviour possible without continually fighting an inherited editor layout and extension model.

The system has one trust boundary

Architecture ownership also reduces the distance between a claim and its implementation. The permission policy sits beside tool execution. The secret store sits beside provider adapters. Model routing can read the same project and task facts as the workspace. Supported WebView and Rust network paths feed the same egress accounting contract.

This does not make the system automatically safe. It makes the system auditable without first separating Miton’s behaviour from millions of lines of inherited product code.

What the choice does not prove

Ground-up architecture is not evidence of speed, quality or cross-platform parity. Those outcomes require measurements and release artefacts. Miton’s approved pre-release evidence shows the application and shared project surfaces. Signed installers and equivalent release outcomes across macOS, Windows and Linux remain v1 gates.

The choice is therefore strategic, not ceremonial. Miton gives up the fastest route to a familiar demo so the project—not the editor, provider or prompt—can remain the unit of work.