1.6.6

What's new

• Multiple QuickAsk turns per message — You can now carry on more than one QuickAsk follow-up from a single assistant reply. When multiple turns exist, a numbered tab bar appears so you can switch between them, and a + button lets you add another turn without closing the panel.
• Markdown in QuickAsk replies — Assistant replies inside the QuickAsk panel are now rendered as proper markdown instead of plain text, so code, tables, and formatting display as intended.
• Insights auto-save — The insight-deepening prompt now instructs the model to save sufficiently complete output directly into the vault via create_note (using the proposed title and tags), and to present it for review only when the content still needs work.

Refinements

• Localized batch_replace_text triggers — The batch-replace tool now carries locale-specific trigger phrases across all supported languages, so the model is more likely to reach for it during non-English conversations. The replace_text and batch_replace_text descriptions were also sharpened to clarify their use cases.
• Broader tool retrieval — The default tool filter now surfaces more candidate tools per turn (top-k raised from 8 to 12), giving the model more options when selecting the right vault tool.
• Modern build targets — The bundle is now compiled to ES2021 (up from ES2018) with a matching ES2022 lib, enabling newer language features in the output.