0.3.5

Built and attested by CI. See CHANGELOG.

0.3.4

Built and attested by CI. See CHANGELOG.

0.3.3

Removes the inline eslint-disable directives that the Obsidian community review disallows. The two unavoidable cases — the chat SSE fetch (requestUrl can't stream) and the deprecated display() kept for minAppVersion 1.7.2 — are handled in the local ESLint config instead. README install section cleaned up.

Assets: main.js, manifest.json, styles.css.

0.3.2

Code-quality patch: all ESLint findings from the community review are now resolved.

• Settings tab's plugin reference is typed (no more any).
• All network requests except the chat SSE stream use Obsidian's requestUrl (CORS-free, mobile-friendly) via a single http module; the stream keeps fetch for incremental token streaming.
• Removed redundant type assertions, voided fire-and-forget promises, dropped deprecated setDynamicTooltip.
• Added lint/typecheck npm scripts (ESLint + typescript-eslint + eslint-plugin-obsidianmd).

Assets: main.js, manifest.json, styles.css.

0.3.1

Patch release addressing the Obsidian community automated review.

• Replaced a direct style.height assignment with setCssStyles.
• Removed the inert declarative settings API so the plugin no longer uses Obsidian 1.13-only APIs (settings render via the classic path on all supported versions).
• minAppVersion bumped to 1.7.2 (revealLeaf is @SInCE 1.7.2); floating leaf promises voided; deprecated setDynamicTooltip and redundant casts removed.
• Manifest/README identity is now fully consistent (vault-retrieval / "Vault Retrieval").

Assets: main.js, manifest.json, styles.css.

0.3.0

Focused RAG core: related notes, semantic search, and grounded local chat.

Highlights in 0.3.0

• Thinking on/off — suppress a reasoning model's thinking via cross-server-portable hints, with a settings test that reports whether the model honours it.
• Model capability hints — best-effort vision/thinking detection (native metadata → name heuristics → live confirmation).
• Embedding model dropdown + inline endpoint test (notice-based).
• Multi-line chat input (auto-grow, Enter/Shift+Enter configurable, IME-safe).
• Context budget ↔ model window coupling.
• Settings layout reworked toward Obsidian's native conventions (incl. forward-compatible Obsidian 1.13 declarative API, classic renderer as fallback).

Note: Image transcription (IMG→MD) moved to the sibling plugin image-to-markdown.

Full changelog: https://github.com/johannes-kaindl/vault-rag/blob/main/CHANGELOG.md

0.2.0 — IMG→MD-Sidebar

0.2.0 — IMG→MD-Sidebar

First published release. Bundles the previously-unreleased features (visible thinking in chat, copy answers, chat model UX) with the new interactive IMG→MD sidebar:

• Checkbox list of the active note's images — select/deselect-all; unsupported formats (HEIC/HEIF) shown disabled
• Per-image live-streaming transcription into result cards, with an optional thinking block and a copy button
• Deliberate writing — "Create note" per card or "Create all" at once (batched, single source-note write); a re-scan after writing drops the handled image from the list
• Built on a shared SSE transport (streamSSE, reused by chat and vision) and a shared batched writer (writeTranscripts, reused by the sidebar and the IMG→MD command); the command and editor context-menu remain

Plus the 0.1.0 foundation: offline related-notes, live indexing, semantic search, and grounded RAG chat.

Canonical home: https://codeberg.org/jkaindl/vault-rag — this GitHub repo is a mirror.

Install (manual)

Download main.js, manifest.json, and styles.css into <vault>/.obsidian/plugins/vault-rag/, then enable the plugin.

See CHANGELOG.md for the full list.