Veynrel

Your notes. Deeper connections.

A knowledge-health and semantic exploration workspace for Obsidian.

See what needs attention, understand your vault's structure, discover relationships you have not explicitly linked, and remember what you learn. Start with local Health and native spaced repetition; enable AI and external integrations when you need them.

One workspace

Health | Findings | Discover | Recall | Connect | Tools | Settings

Page	What you can do
Health	See Vault Pulse, recommendations, four dimensions of knowledge health and Vault Topology.
Findings	Review evidence and decide what to address, dismiss or revisit.
Discover	Search by meaning, explore semantic maps and inspect connection candidates.
Recall	Review your flashcards with native FSRS-6 scheduling.
Connect	Configure Companion, synchronize its mirror and review external proposals.
Tools	Launch advanced audits, Ask your Vault, batch processing and MOC generation.
Settings	Check capability state and configure the services you use.

A look inside

Synthetic vault data, shown in native dark and light themes. Select an image to inspect it at full size.

Health — attention, evidence and next steps.	Global Semantic Map — explore your indexed knowledge.
Connection Opportunities — inspect relationships alongside your links.	Recall — a focused review, inside Obsidian.

Quick start

Install Veynrel, then use its ribbon icon or Open Veynrel Health command.
Choose a profile and run local Health. Inspect the recommendation and Findings; no AI provider is required.
Optionally enable Semantic Intelligence: choose an embedding provider/model, test the connection and explicitly build the first index.
Open Discover to explore Semantic Neighborhood, Global Semantic Map and Connection Opportunities.
If you want spaced repetition, add flashcards and choose Find flashcards in Recall.
Configure Deep Intelligence for Knowledge Health or advanced AI workflows, and Connect for Companion, only if needed.

Health and Findings

Vault Pulse → recommendation → current data. Health shows concrete Findings and coverage instead of an opaque 0–100 score, across four dimensions:

Structure: broken links, orphan notes, weak structure and disconnected areas.
Connections: explicit relationships, exact duplicates and semantic duplicate findings.
Recall: cards due for native review.
Knowledge: explicit AI-assisted signals about underdeveloped notes.

The current-data dashboard brings findings, last-check information, Recall and Knowledge state together. Vault Topology lets you inspect actual Markdown links in a map with note details. The local Health scan requires no AI.

Findings is your action inbox. Review evidence, open affected notes, dismiss, snooze or reopen a Finding. Supported checks automatically resolve findings when a subsequent analysis confirms the issue has disappeared.

Discover by meaning

Search & find: semantic search, related notes and potential semantic duplicates.
Visual exploration: Semantic Neighborhood, Global Semantic Map with selected-note focus, and Connection Opportunities.
Semantic Health: explicitly check for semantic duplicates and review the resulting Findings.

These features reuse the existing semantic index. Similarity is a discovery signal, not proof that notes are identical or agree.

Relationship view	What it shows
Vault Topology (Health)	The Markdown links you actually created.
Semantic Neighborhood	One note and its closest semantic neighbors.
Global Semantic Map	The whole eligible indexed semantic space, within the current map limit.
Connection Opportunities	Semantic top-5 relationships compared with explicit Markdown links.

Global Semantic Map

Distance from the center reflects semantic similarity to the current center: closer means more similar. The default center is the vault semantic core; any mapped note can explicitly become the focus, with a simple reset to the core.

Node size reflects semantic connectedness, not importance or note quality. Map exploration reuses existing embeddings and makes no new embedding requests. See the map guide for details.

Connection Opportunities

Inspect connection candidates with semantic evidence beside the Markdown relationship:

Category	Meaning
Candidate	A semantic top-5 relationship with no known Markdown link in either direction.
Aligned	A semantic top-5 relationship that also has Markdown linkage.
Explicit-only	A Markdown-linked pair outside the sparse semantic top-5 graph.

Explicit-only does not mean semantically unrelated. Candidates are invitations to inspect, not proof that a link should exist. Coverage notices identify relationships that could not be compared.

Review aids include mutual top-3, mutual top-5 and one-sided evidence; shared semantic neighbors; and an optional Hide Excalidraw notes presentation filter. Mark a candidate Useful / Not useful / Unsure during the current session. These annotations are not persisted, and no Markdown links are created automatically.

More: Connection Opportunities · candidate evidence and review.

Remember with Recall

Native spaced repetition uses FSRS-6, a due queue and local scheduling. No third-party spaced repetition plugin is required.

Write one Question::Answer card per line under a Markdown heading named exactly Flashcards (case-sensitive):

## Flashcards

What is retrieval practice::Actively recalling information instead of rereading it
Why use spaced repetition::It schedules reviews near the point of forgetting

Choose Find flashcards, start a review, reveal the answer and rate Again / Hard / Good / Easy (keys 1–4). Schedules survive restarts.

AI card generation is optional and explicit: it uses your configured language model and appends generated cards to the note for native Recall. Recall guide.

Connect, Tools and Settings

Connect works with the optional, self-hosted Veynrel Companion. It makes a synchronized mirror available to MCP clients for note reading, semantic search and change proposals. External agents cannot directly write your vault through the proposal workflow: you inspect each proposal in Obsidian and choose Approve or Reject.

Manual mirror sync requires confirmation. Enabling Connect also permits disclosed incremental synchronization after subsequent semantic changes. The mirror can contain Markdown, chunk text, metadata and embeddings; use an endpoint you trust. Connect data flow and setup.

Tools keeps Ask your Vault, Deep Audit / Single Audit, batch processing, MOC generation and legacy reports available. Editor workflows—AI writing, selection transforms, Dataview generation and atomization—remain separate, through the command palette and editor context menu. Existing command IDs remain compatible with hotkeys and automation.

Settings summarizes Deep Intelligence, Semantic Intelligence, Connect and Recall. Native Obsidian Settings → Veynrel provides advanced provider/model, embedding, Companion, Deep Audit, output and interface controls.

Supported language-model providers include Ollama, OpenRouter, OpenAI, Groq and custom OpenAI-compatible endpoints. Embeddings support Ollama, OpenRouter and OpenAI-compatible endpoints. Language-model, embedding and Companion credentials are separate.

Local-first, explicit AI

Activity	Where the work happens
Local Health and Vault Topology	On your device; no AI provider required.
Recall discovery and scheduling	Locally, using native FSRS-6.
Semantic indexing / search	A remote embedding provider receives required note chunks / the search query. A local Ollama endpoint can keep embedding work local.
Semantic maps / connection comparison	Existing index and link metadata; no new embeddings or provider calls.
Knowledge Health / AI workflows	Explicit actions send the content needed for the task to your configured language model.
Connect	The disclosed mirror goes to your configured Companion endpoint; proposed vault changes require Obsidian approval.

Opening the workspace does not itself scan, build an index, call a provider or synchronize Companion. The first semantic build is explicit; later Markdown edits can update an existing index incrementally. A local vector index does not make a remote provider local.

Veynrel has no telemetry or analytics. Provider API keys are not sent to Companion, and the Companion token is not sent to AI providers. Review the policies of remote services before sending sensitive notes. Data-flow, storage and recovery details are in docs/.

Installation

Obsidian Community Plugins

Open Settings → Community plugins → Browse.
Search for Veynrel.
Install and enable it.

Manual installation

Download exactly these three files from the same GitHub Release:

main.js
manifest.json
styles.css

Place them in:

<your-vault>/.obsidian/plugins/ai-knowledge-hub/

Reload Obsidian and enable Veynrel.

Updating

The plugin ID remains ai-knowledge-hub. Existing users update normally; do not create a second veynrel plugin folder. Compatible settings and feature data continue to use the existing plugin directory.

How it fits together

Obsidian vault
├─ Local Health → Findings
│                Vault Topology
├─ Semantic Index → Search / Related / Duplicates
│                   Semantic Neighborhood
│                   Global Semantic Map → Selected-note focus
│                   Connection Opportunities ← Markdown links
├─ Recall → FSRS-6
├─ Deep Intelligence → Knowledge Health / advanced AI workflows
└─ Connect → Companion / MCP → proposals → approval in Obsidian

Analysis may suggest. Veynrel shows evidence. You decide.

Current limitations

The first semantic build and rebuilds after incompatible embedding-space changes are explicit.
Global Semantic Map exact mode supports up to 500 eligible mapped notes; Connection Opportunities shares that limit.
Connection Opportunities uses a sparse top-5 semantic graph. It cannot prove a Markdown link should exist; candidate reviews are session-only, not persisted.
Recall has no decks, review-history views or optimizer analytics.
Semantic search uses a local linear scan; duplicate detection uses pairwise comparison suited to personal vault sizes.
Ask your Vault is a one-shot flow, and Knowledge Health is an AI-assisted review signal, not factual verification.
Companion remains self-hosted, with no hosted Veynrel account service.
Verification of the newer visual/native surfaces is strongest on Linux desktop. Mobile and other desktop environments are not equally covered for every surface.

Development and contributing

npm ci
npm test
npm run typecheck
npm run lint
npm run audit:proposals
npm run build

Veynrel and Companion are separate repositories. For integration development, clone Companion beside this repository as ../veynrel-companion, then run npm run companion:smoke-sibling.

See docs/ for architecture and storage contracts, or the final UI/UX audit for native verification and known interface limitations. Issues, bug reports and pull requests are welcome.

Support Veynrel

Veynrel is free and open source. Star the repository or support development on Boosty.

veynrel