Starstruk Logo MarkStarstruk

v1.10.0

hkchengrex/MMAudiov1.10.0Jun 18, 2026by lfnovo

AI Summary

This release introduces LaTeX math rendering in chat, bulk controls for chat contexts across 14 locales, and Turkish localization. It also addresses security vulnerabilities, improves ingestion and search reliability, and includes a database migration for flexible configuration.

Key Highlights

  • LaTeX math rendering in chat via KaTeX
  • Bulk chat-context actions for Sources and Notes columns
  • Turkish (tr-TR) UI localization
  • Security fix for CVE-2026-48710 via dependency updates
  • Reliability fixes for ingestion and search processes

New Features

  • LaTeX math rendering in chat
  • Bulk chat-context actions
  • Turkish UI localization
  • Configurable frontend API timeout
  • OpenRouter embedding modality exposure

Full Release Notes

Highlights: LaTeX rendering in chat, per-type bulk chat-context controls, Turkish UI, a security fix, and a batch of ingestion/search reliability fixes.

> **Upgrade note:** this release adds database **migration 15** (a flexible `config` object on the `credential` table), applied automatically on API startup.

## 🔒 Security
- Bumped **Starlette → 1.2.1** and **FastAPI → 0.136.3** to address **CVE-2026-48710** ("BadHost") (#859)

## ✨ Added
- **LaTeX math rendering in chat** — inline `$...$` and display `$$...$$` via KaTeX (#606)
- **Bulk chat-context actions** in the Sources and Notes column headers — sources: *insights only* / *full content* / *exclude all*; notes: *include / exclude all* — translated across all 14 locales (#223)
- **Turkish (tr-TR)** localization (#871)
- `NEXT_PUBLIC_API_TIMEOUT_MS` to configure the frontend API request timeout (#880)

## 🔧 Changed
- Prominent **"Retry processing"** button on failed source cards (#726)
- Docker base image → **Debian trixie + Node.js 22.x** (#914)

## 🐛 Fixed
- Sources that fail to ingest are now marked **`failed`** (so the retry button actually appears) instead of being saved as `completed` with the extraction error as their body (#726)
- Podcast generation now uses the notebook's real content via `Notebook.get_context()` (#864)
- `PUT` profile handlers use `model_dump(exclude_unset=True)` so partial updates don't wipe unset fields (#860)
- OpenRouter embedding modality is now exposed (#842)
- `POST /sources/{id}/retry`: no longer returns `400` for every source (#861), and no longer `500` from a double-prefixed command id
- `GET /sources/{id}` returns `404` (not `500`) for a missing or deleted source
- Text search falls back to vector search on a `search::highlight` "position overflow" (#648)
- `POST /api/search` rejects a non-positive `limit` with `422` (#863)
- Ollama `num_ctx` override is now persisted via a flexible `config` object (migration 15) (#875)
- Worker drains legacy embedding jobs via command aliases (#695, #876)
- Updated AI provider configuration docs (#873)

## ⚡ Performance
- Memoized the notebook source list and stopped polling completed sources (#503)

## 📚 Docs
- Code of Conduct (#856), local quickstart + worker step (#900), `export_docs.py` Table of Contents (#865)

## 📦 Dependencies
- Multiple dependency bumps, including **cryptography 46→48**, **langchain 1.2→1.3**, langchain-anthropic, aiohttp, tornado, and frontend dev dependencies.

## 🙌 Contributors
Thanks to everyone who shipped a PR in this release: @aaronjmars, @doganreis, @Federicorao, @fboudra, @krataratha, @ltianyi992, @mvanhorn, @royfrancis, @zichen0116, and @lfnovo.

**Full Changelog**: https://github.com/lfnovo/open-notebook/compare/v1.9.0...v1.10.0