debe
released this
2026-10-12 08:13:17 +00:00
| 11 commits to main since this release
Fixed
- JSON extraction is now robust across models. A local Qwen build returns an extra
opening brace ({{"summary_ar": ...}) and reasoning models prepend a thinking block;
both made every analysis fail with "Could not parse the JSON from the AI response".
The extractor now strips thinking blocks and markdown fences, tries the whole string
first, then scans for the first balanced {...} span that actually parses —
ignoring braces inside string values.
- An empty model response now reports "The AI returned an empty response" instead of a
generic parse error.
Added
- Five regression tests for the extractor: extra opening brace, thinking block, prose
around the JSON, braces inside strings, and the empty-response case.
Downloads