0.8

[Chenglong Wang (Chenglong-MS)]

This pull request introduces a portable desktop build for Data Formulator, adds ClickHouse support, improves Azure CLI integration, and strengthens model config security. The most significant changes are summarized below.

---

Desktop Application Support

• Added a new portable desktop build using PyInstaller and pywebview, allowing users to run Data Formulator as a standalone native app without needing to install Python or Node.js. Includes a new entrypoint (data_formulator_desktop.py), build script, and PyInstaller spec file. [1] [2] [3] [4] [5] [6]
• The desktop launcher ensures the Azure CLI is available in PATH by searching common install locations and amending the environment as needed. [1] [2]

Azure CLI Integration and Authentication

• Refactored Azure CLI detection and invocation into a reusable utility (expose_azure_cli), used by both the backend and desktop launcher to ensure seamless authentication for Kusto and other Azure connectors. [1] [2] [3] [4] [5]
• When running as a desktop app, Azure authentication uses AzureCliCredential instead of DefaultAzureCredential for improved user experience. [1] [2]
• Suppressed browser auto-launch for the backend server when running in desktop mode.

Database Support

• Added ClickHouse support as a data source, including dependency, loader registration, and SQL dialect definition. [1] [2] [3]

Security Improvements

• Hardened model config handling: the is_global flag is now validated by checking the server registry rather than trusting client claims, preventing privilege escalation and SSRF attacks. [1] [2]
• Improved error handling for API base URL allowlist validation.

Other Improvements

• Changed SQL Server data loader to use SELECT TOP n for limiting results, simplifying query construction. [1] [2]
• Added auth_mode to the public model info to distinguish between Azure Identity and API key authentication.
• Minor frontend refactor: added useRef and useNavigate imports in App.tsx. [1] [2]