Studio installation and workspace guide

ZBL Studio AI is an offline-first desktop IDE. Its Tauri 2 window hosts a React, TypeScript, Vite and Monaco workbench based on the supplied neon layout. Rust provides scoped file operations, workspace trust, native ConPTY or PTY sessions, fixed Zebvix command dispatch, Git status and OS credential storage.

Windows preview

The v0.3.2 per-user QA installer includes the manifest-matched v0.5.0-preview.5 Zebvix binary. WebView2 is required. An external compatible Clang is required for native ZBL builds. Interpreter and bytecode run from the bundled CLI.

1. Open the setup executable and follow the installer. Launch ZBL Studio AI from the Start menu.
2. Create a ZBL project or open a folder. The File menu also accepts an absolute workspace path. The Explorer opens files and supports create, rename, Save As and deletion of files or empty folders.
3. Trust a workspace only when you intend to allow local processes. Choose Bytecode or Interpreter, then use Check, Build, Run or Test. The terminal also offers an interactive PowerShell tab.
4. To remove the IDE use Windows Installed apps. Workspaces are stored where you chose them and are preserved by the uninstaller.

Developer intelligence preview

The v0.2 QA build includes the standalone zbl-language-server and uses its compiler-backed query engine in Monaco through a scoped Tauri bridge. Open a ZBL source file for diagnostics, completion, hover, signature help, references, definition navigation and formatting. The View menu opens the document outline, searches the opened project graphs and previews a private-task rename. Applying the preview changes the editor buffer. Save writes it to disk.

For interpreter debugging, trust the workspace, save the source, set a breakpoint in the gutter and choose Run → Debug project or Debug current file. The Debug panel shows call frames and the innermost frame's locals when execution pauses. Build, Run and Test still use the bundled Zebvix SDK. The v0.2 capability and QA matrix explains the supported subsets. Developer intelligence guide

Local contract simulation

Create a Counter contract from the welcome screen or command palette. Open Run → Simulate local contract in a trusted workspace, then use Build artifact to compile saved project source through the bundled CLI. You can also browse to a .zct file inside the active workspace. Enter an initial state and a comma-separated list of inputs. The workbench shows the artifact SHA-256, the before/after timeline, and the canonical JSON receipt. A failed step reports its position. Studio and zebvix contract simulate counter.zct 40 2 -1 1 use the same offline ZBR engine. ZBX chain execution is a separate target.

Expand Replay receipt to save a versioned receipt to a new workspace file or verify a saved receipt by replay. Receipt commands and limits

Use Ctrl+K for the command palette and Ctrl+P for workspace search. In Tools, choose Graphite, Light, or High Contrast and adjust panel visibility. Studio remembers these preferences and panel sizes. The Git panel shows actual staged and working tree patches from the trusted workspace.

In the Git panel, open a file from the active repository, then choose Stage current file or Unstage current file and confirm. Staging uses the saved file. Unstaging updates its index entry while keeping the saved file and editor buffer. These controls require a trusted workspace with a local .git directory and refuse unresolved conflicts.

AI and security

Manual mode uses the local SDK without an account or subscription. OpenAI BYOK is opt-in and uses an OS-stored key. Sending chat sends the prompt and optionally the active file. Requesting an edit always sends the active file and displays a Monaco diff. Apply changes the editor buffer, and a separate Save action writes it. Studio never invokes a process based on model output.

The file bridge rejects traversal, alternate data streams, and reparse points within a chosen workspace. A trusted shell can use the account's ordinary operating system permissions. Workspace trust is session-scoped. Filesystem checks assume a stable workspace and are not an operating-system sandbox.

Build and verify on Windows

Use Node 24, Rust and a Windows C/C++ toolchain. The internal candidate was built using the existing Windows GNU toolchain. Tauri's standard Windows workflow uses MSVC. Use the manifest-matched SDK CLI from the existing SDK build.

./studio/scripts/prepare-windows.ps1 -Manifest releases/v0.5.0-preview.5.json -Artifacts $sdkArtifacts
./studio/scripts/prepare-language-server.ps1
Set-Location studio
npm ci
npm run tauri -- build --config src-tauri/tauri.developer-qa.conf.json --target x86_64-pc-windows-gnu --bundles nsis -- --locked

Set ZBL_STUDIO_TEST_PRODUCT to ZBL Studio Developer QA, ZBL_STUDIO_TEST_SDK_MANIFEST to the preview.4 manifest path, and ZBL_STUDIO_TEST_SIMULATION to 1. Run node studio/scripts/verify-windows.mjs <setup.exe> <evidence-directory> from the repository root. The script installs to an isolated location, uses the real WebView UI and removes the test installation afterward. Its loopback WebView2 DevTools listener runs only for verification. The QA setup uses its own per-user application registration and Start Menu folder.

For the v0.3.1 installer set ZBL_STUDIO_TEST_VERSION=0.3.1 and ZBL_STUDIO_TEST_GIT_INDEX=1 to verify the single-file index workflow.

For the v0.3.2 installer select the preview.5 manifest, set ZBL_STUDIO_TEST_VERSION=0.3.2 and enable both ZBL_STUDIO_TEST_GIT_INDEX=1 and ZBL_STUDIO_TEST_RECEIPTS=1.

Use CARGO_BUILD_JOBS=1 if the Windows build reports paging-file error 1455. The v0.3 acceptance matrix records the scope of this internal preview.

Portal

Run the Next production server bound to loopback with ZBL_STUDIO_LOCAL_PREVIEW=1 and ZBL_STUDIO_LOCAL_RELEASE_DIR set to the absolute verified artifact directory. Visit /studio for the verified setup link. Normal production configuration leaves Studio binary delivery disabled. Public deployment and signing require release approval.

Scope

External language-server adapters, standard DAP transport, Git clone and write operations, hosted accounts, managed subscriptions, multi-agent orchestration, teams, extension installation and ZBVM contract execution remain pending. Monaco supplies the existing syntax support and built-in web-language services. Linux packages and runtime require a Linux WebKitGTK build and test environment. Provider requests require a usable OpenAI key and model access.