Skip to content

Releases: slow-coding/obsidian-llm-content-block

0.1.1 — review fixes

Choose a tag to compare

@slow-coding slow-coding released this 10 Jul 19:04

0.1.1

Fixes from the Obsidian plugin review (same features as 0.1.0):

  • Settings heading via Setting.setHeading() instead of raw <h3>
  • Textarea styles via CSS class instead of inline static styles
  • Type-safe settings load (loadData() cast)
  • Removed unnecessary regex escape
  • Caught a floating promise in the render loop
  • Release assets trimmed to main.js / manifest.json / styles.css

Manual install

Download the 3 files into <vault>/.obsidian/plugins/llm-content-block/, then enable in Settings → Community plugins.

0.1.0 — initial release

Choose a tag to compare

@slow-coding slow-coding released this 10 Jul 18:47

LLM Content Block 0.1.0

Paste AI conversations into Obsidian as scrollable, format-safe cards — their markdown stays contained, your notes stay clean.

Features

  • aicard code block → scrollable glass card
  • Role-aware bubbles (user / AI); source pill with 12 preset models (GPT / Gemini / Claude / DeepSeek / Qwen / ERNIE / Kimi / Doubao / Gemma / GLM / Grok / Llama) + custom
  • Full markdown / LaTeX / code support; long AI answers collapse by default
  • In-place edit modal; "Insert AI chat block" command + hotkey
  • Theme-aware (CSS variables, dark/light auto); i18n (follows Obsidian locale)

Manual install

Download manifest.json, versions.json, main.js, styles.css into:
<vault>/.obsidian/plugins/llm-content-block/
Then enable it in Settings → Community plugins.