Release notes · v0.37.1
Released on 2026-10-12 UTC. Official release.
Preserve scientific PDF structure
Figure extraction better preserves captions, legends and multi-panel graphics. Ruled tables beside formulas retain grouped headers, row structure and source-cell references. Captions, graphics and text are kept with their source element rather than a neighboring one.
When using Figures & Tables, compare the extracted element with its original page before citing it. Check complete panels, legends, table headers, units and footnotes. Translated PDF regeneration also accepts valid differences in embedded glyph names while preserving the original source ink; use the comparison and export checks.
Windows and installation
- The desktop CLI handles installation paths containing Chinese and other non-ASCII characters, preserving the caller's code page and exit code. Update and reopen the terminal if an older launcher misreads its path; see CLI setup.
- Local PDF translation handles Windows drive paths correctly.
- Notebook kernels, environment fingerprints, package inventories and lock capture avoid flashing extra console windows.
- Desktop and standalone runtime packages remove unused duplicate dependencies and assets, reducing the download growth introduced in v0.37.0.
The standalone CLI and Node backend introduced in v0.37.0 remain the basis for service operation. Use the package matching your platform and keep the extracted runtime directory intact.