v1.3.1

spider-rs/spiderv1.3.1Feb 6, 2026by denizsafak

AI Summary

A bug fix release improving the JSON parsing logic in the automation module to handle LLM responses that include reasoning text before the JSON code block.

Key Highlights

Improved `best_effort_parse_json_object` parsing function
Ability to find JSON blocks anywhere in the response, not just at boundaries
Support for JSON arrays in addition to objects
Better fallback parsing for various LLM response formats

New Features

Robust parsing of LLM reasoning text
Handling of JSON arrays in extraction results
Flexible fallback parsing for different model outputs

Full Release Notes

- Special thanks to [@jeremiahsb](https://github.com/jeremiahsb) for his [massive contribution](https://github.com/denizsafak/abogen/pull/120) (>55k lines!) that brought the Web UI, EPUB 3 pipeline, and core architectural improvements to life.
- Added an EPUB 3 packaging pipeline that builds media-overlay EPUBs from generated audio and chunk metadata.
- Persisted chunk timing metadata in job artifacts and exercised the exporter with automated tests.
- Added Flask-based Web UI (`abogen-web`) for Docker and headless server deployments.
- Reorganized codebase to support both PyQt6 desktop GUI and Web UI from a shared core.
- Added Supertonic TTS engine support with GPU acceleration.
- Added entity analysis and pronunciation override system for proper nouns.
- Added speaker/role assignment for multi-voice "theatrical" audiobooks.
- Added Calibre OPDS and Audiobookshelf integration.

Update:
```
uv tool update abogen
```