Skip to main content

Known Issues

Current limitations, platform quirks, and workarounds.

This is a living document. Things get fixed, new issues appear. Check GitHub Issues for the most current list.

Platform-Specific​

Linux​

X11 support

  • Status: Not tested, not a priority
  • Workaround: Use XWayland or native Wayland
  • Notes: Most modern Linux distros default to Wayland now

Windows​

SmartScreen warnings on install

  • Status: Release builds are signed with Azure Trusted Signing. A new release can still trigger SmartScreen until it builds reputation.
  • Workaround: Click "More info" then "Run anyway"

Path length limits

  • Issue: Windows has a 260 character path limit (unless registry-enabled)
  • Workaround: Keep workflow files close to root (e.g., C:\Data\ not C:\Users\...\OneDrive\...\)
  • Notes: Rare in practice, but affects deeply nested project structures

macOS​

Gatekeeper/notarization

  • Issue: macOS blocks apps from "unidentified developers"
  • Root cause: No Apple Developer certificate yet
  • Workaround: Right-click -> Open -> Open anyway
  • Future: Notarization is on the roadmap

Apple Silicon (M1/M2/M3)

  • Status: Release builds target x64 only. They run on Apple Silicon through Rosetta 2.
  • Notes: Report issues if you encounter them

Tool-Specific​

Filter Tool​

Complex nested expressions

  • Issue: Very deeply nested Polars expressions can be slow to validate
  • Workaround: Break complex filters into multiple Filter tools
  • Notes: Execution is fine, it's just the real-time validation

Formula Tool​

No autocomplete for chained methods

  • Issue: Autocomplete shows .str. methods but not what's after
  • Example: pl.col("name").str. suggests methods, but pl.col("name").str.contains() doesn't suggest .alias()
  • Workaround: Refer to Polars docs for method chains
  • Notes: Monaco autocomplete is tricky to extend

Join Tool​

Many-to-many join performance

  • Issue: Joining two large tables on non-unique keys can explode row count
  • Notes: This is how joins work, not a bug, but the UI doesn't warn you
  • Future: Add row count estimates before execution

Input Tool​

Large file preview delays

  • Issue: Opening a multi-GB file takes noticeable time for initial preview
  • Root cause: Need to scan file for schema inference
  • Workaround: Be patient, or use smaller sample files during development
  • Notes: Execution uses lazy evaluation, so the actual pipeline is fast

General​

Undo/redo across tabs

  • Issue: Undo history is per-tab, but it can feel disorienting when switching
  • Notes: This is by design (each tab is independent), but some users expect global undo

Large workflows (100+ nodes)

  • Issue: Canvas performance degrades with very large workflows
  • Root cause: React re-renders, Xyflow limitations
  • Workaround: Break large workflows into smaller sub-workflows
  • Future: Investigate virtualization and rendering optimizations

Memory usage

  • Issue: Electron + Chromium + bundled Python can use significant RAM
  • Typical: 500MB-1GB for normal use
  • Peak: Can spike higher during large data operations
  • Notes: This is the Electron tax. Desktop apps have memory to spare.

Browser Mode (Development Only)​

The app can run in a browser for frontend development, but with limitations:

No native file access

  • Issue: Browser security prevents reading/writing local files
  • Workaround: File picker works, but paths may not resolve correctly
  • Notes: Use Electron mode for actual workflows

No backend bundling

  • Issue: Browser mode connects to a separate backend process
  • Notes: Fine for UI development, not for testing full workflows

Reporting New Issues​

When reporting issues:

  1. Check existing issues first
  2. Include environment: OS, version, architecture (x64/ARM)
  3. Steps to reproduce: What exactly you did
  4. Expected vs actual: What should have happened vs what did
  5. Workflow file: If relevant, attach (sanitize sensitive data first)
  6. Logs: Check DevTools console (Ctrl+Shift+I)

Good bug reports get fixed faster.

Security Notes​

Workflow trust boundaries

  • Formula and Filter expressions use a restricted Polars namespace. File and network readers, Python callables and disallowed attribute chains are rejected before evaluation.
  • On a single-user desktop, the workflow author controls the machine. Native file access still requires an open/save dialog grant; use trusted workflows and keep backups.
  • In Hub mode, workflow authors are untrusted relative to the Server. Expressions must not bypass Hub-issued connection or file grants. Use Hub file-access mode and restricted database accounts on Servers you control.
  • The expression sandbox does not isolate the whole Server process or guarantee resource isolation. See the Security Model for the limits.

Data stays local

  • Desktop processing runs on the local Server; Hub workflows run on the selected Server.
  • Database connections and configured AI providers can receive requests and data. AI use sends workflow context and sample data to that provider.
  • Your data, your responsibility: Back it up, secure it appropriately

Next: Return to Developer Guide or explore Architecture.