Veynrel 1.8.0

Veynrel 1.8.0

Vault Audit AI is now Veynrel.

This is the same Community Plugin with the same plugin ID, settings, local data, semantic index, and update path. Existing users do not need to reinstall anything.

No action required for existing users. Update normally. Your API keys, provider choices, hotkeys and existing semantic index stay in place; the new name does not require reindexing. Obsidian 1.8.7 remains the minimum supported version.

A name for the whole workflow

Veynrel brings together semantic search, related-note discovery, potential duplicate detection, Ask your Vault, AI-assisted writing and vault auditing. The README and settings now reflect that wider workflow. Semantic features and the optional Companion remain opt-in.

Safer work with your notes

• AI replacements and flashcard writes check for edits made while generation was running. Batch backups preserve relative note paths, and failed backups stop replacement.
• Simultaneous requests to atomize the same note share one operation.
• Streamed responses preserve their final tokens, and empty AI output is rejected.
• Vector validation rejects values that cannot be stored safely, protecting the local semantic index.
• AI change proposals can be inspected as a text diff and explicitly approved or rejected in Obsidian. A proposal alone never changes a note; stale or conflicting targets are rejected.

Optional Companion and MCP

Veynrel Companion can mirror semantic state to a local server or your own HTTPS endpoint. It offers scoped MCP retrieval and change proposals, with optional Qdrant search acceleration. Obsidian remains the only writer to your vault, and provider credentials are separate from Companion credentials.

The standalone server now lives in its own repository. Plugin integration remains included. The repository name and existing configuration, environment variables and protocol remain compatible; existing deployments need no reconfiguration for this rebrand. See the setup and privacy documentation before enabling synchronization.

Settings and installation

• Obsidian 1.13+ can find plugin options through its settings search. Older supported versions keep the existing settings interface.
• Community Review cleanup improves note-write safety and use of supported Obsidian APIs.
• Dependency installation is compatible with the npm 10 and npm 11 versions used in scanner checks.

Veynrel remains free and open source. Optional support is available on Boosty.

Vault Audit AI 1.7.0

Vault Audit AI 1.7.0

This release adds Ask your Vault, a source-grounded RAG workflow built on top of the existing local semantic index.

What's new

• added Ask your Vault for one-shot Q&A over indexed notes
• answers stream progressively and are grounded in selected Vault sources
• added trusted source cards with exact note paths and best-effort source navigation
• added multi-document context selection with bounded context and document diversity
• reconstructs current Markdown chunks instead of using shortened vector-index previews
• detects stale or changed indexed sources before sending context to the model
• keeps RAG context frozen during generation so AutoSync, Clear, and Rebuild remain safe
• added prompt-injection boundaries for retrieved note content
• validates model citation IDs against code-owned trusted sources
• added cancellation and safe modal-close behavior
• hardened the Ask your Vault modal for long answers, long paths, many sources, and smaller screens
• final completed answers now render as native Obsidian Markdown

Privacy

Ask your Vault performs one semantic query embedding. If the embedding provider is remote, the question is sent to that provider.

If the configured language-model provider is remote, only the question, selected reconstructed source chunks, and required metadata are sent for answer generation.

The full Vault and vector index are not sent to the language-model provider.

Ask your Vault is read-only and never modifies Markdown notes.

Reliability

• 698/698 automated tests passing
• 0 ESLint errors
• TypeScript typecheck passing
• production build passing
• manual Obsidian smoke test completed successfully

Vault Audit AI 1.6.1

Vault Audit AI 1.6.1

This patch focuses on Obsidian plugin health, compatibility, and runtime reliability.

Improvements

• added type-aware Obsidian ESLint checks
• reduced local plugin-health findings to 2 justified warnings
• fixed unsafe TypeScript boundaries
• made plain tsc --noEmit pass
• replaced non-streaming fetch requests with Obsidian requestUrl
• preserved real incremental SSE streaming with full timeout and abort handling
• added runtime validation for external API responses
• hardened semantic store initialization errors
• improved custom Obsidian config-directory support
• removed generated declaration artifacts and unused Node tooling
• switched production bundling to a browser-compatible target

Reliability

• 627/627 tests passing
• 0 ESLint errors
• browser and Safari/mobile-compatible bundles verified
• runtime Node-only dependency scan clean

Two plugin-health warnings remain intentionally:

• fetch is still used for true incremental SSE streaming
• the current dynamic settings UI intentionally remains on PluginSettingTab.display()

Vault Audit AI 1.6.0

Vault Audit AI 1.6.0

This release completes the local semantic discovery MVP.

What's new

Automatic semantic index synchronization

After the initial manual indexing, the semantic index now automatically follows Markdown changes in the Vault:

• created notes are indexed automatically
• modified notes update only changed chunks
• renamed notes move to the new Vault path
• deleted notes are removed from the index
• rapid edits are debounced and coalesced
• compatible indexes are reconciled after Obsidian starts

Initial index creation and incompatible-model rebuilds remain explicit.

Similar Notes

A new Find similar notes command discovers notes related to the active Markdown file using the vectors already stored in the local semantic index.

• no additional embedding request is required
• the current note is excluded
• results are ranked at document level
• the most relevant chunks are shown as context
• selecting a result opens the exact Vault path and relevant location

Potential Semantic Duplicates

A new Find potential semantic duplicates command scans the local semantic index for highly similar notes.

• exact local comparison
• deterministic ranking
• duplicate A/B pairs are shown only once
• short or semantically incoherent documents are excluded
• results are presented as potential duplicates, not automatic conclusions

The plugin never automatically merges, deletes, or rewrites notes.

Privacy

Semantic discovery works entirely on the existing local vector index and does not require additional provider requests.

Automatic indexing may send new or changed Markdown chunks to the configured embedding provider. With a local Ollama provider, embedding generation can remain local as well.

Reliability

This release includes extensive regression, concurrency and adversarial testing around automatic synchronization, Clear/Rebuild lifecycle, committed snapshots, single-writer coordination, document similarity and duplicate discovery.

Vault Audit AI 1.5.0

Vault Audit AI 1.5.0

This release adds a persistent local semantic index and meaning-based search for Obsidian notes.

Added

• Markdown-aware semantic indexing
• Incremental chunk-level updates
• Local persistent vector storage
• Semantic search grouped by note
• Manual indexing for the full Vault and the current note
• Exact-path note opening and navigation to the most relevant section
• OpenRouter, Ollama and OpenAI-compatible embedding providers
• Semantic index status, clear and explicit rebuild controls

Reliability

• Atomic persistence with backup-aware recovery
• Embedding-space compatibility checks
• Concurrency protection for search, indexing, clear and rebuild
• Provider-free status and empty-index checks
• Safe runtime invalidation when embedding settings change
• 499 automated tests
• Manual smoke-testing in a separate Obsidian Vault

Privacy

The vector index and similarity search are stored and executed locally.

When OpenRouter or another remote embedding provider is selected, note chunks and search queries are sent to that provider. Ollama can be used for local embedding generation.

Semantic features are opt-in and disabled by default.

Current limitations

Semantic indexing is manual in this release. Automatic synchronization after creating, editing, renaming or deleting notes is planned for a future version.

1.4.0

1.4.0

Added

• Split a note into atomic notes (Zettelkasten). The "Split into atomic notes" command breaks an overloaded note into one note per idea, each linking back to the original. The original is kept intact and gets a section linking to its atoms.
• By default, atoms are created in the same folder as the source note (preserving topic-based organization). A setting allows sending them to a dedicated folder instead.
• Notes that are already atomic are detected and left untouched.

1.3.0

1.3.0

Added

• MOC (Maps of Content) generation. After running a deep audit, use the "Generate MOC from clusters" command to create a hub note for each topic cluster — with a short AI-generated summary and wikilinks to all notes in the cluster. Output folder is configurable (default: MOCs/).
• Audit clusters are now persisted in the note index, so MOC generation reuses them without re-running the audit (saves tokens).

Changed

• Index schema bumped to version 3. Existing indexes reset on first run after update; the next audit will be a full pass (one-time).

1.2.2

Up minAppVersion

1.2.1

1.2.1

Fixed

• Language detection now uses Obsidian's getLanguage() API instead of reading localStorage.
• Replaced a remaining setTimeout with window.setTimeout for popout window compatibility.
• Minor type-safety cleanups in the note index.

1.2.0

1.2.0

Added

• Flashcard generation for the Spaced Repetition plugin. Generate question::answer cards from a note's content — available both as a batch preset and as a command / editor menu action for the current note.

Note

• Flashcards work best with mid-size or larger models; very small free models may copy the prompt examples instead of generating cards.

Fixed

• Filter section layout in the batch modal (stray divider and column misalignment).