Vault Steward: Keep large Obsidian vaults trustworthy

Disclaimer

Is this plugin open source? Yes
Is this plugin completely free? No. Vault Steward itself is free, and local providers such as Ollama can be used without API charges, but optional cloud model providers may incur third-party costs.
Is this plugin vibe-coded beyond the author’s ability to comprehend how it works? No
Community Directory: https://community.obsidian.md/plugins/vault-steward


Vault Steward: Keep large Obsidian vaults trustworthy

As an Obsidian vault grows, maintaining it becomes a different problem from organizing it.

Broken references accumulate. Frontmatter drifts. Old decisions remain beside newer ones. Tasks become detached from the decisions that created them. Notes can contradict each other while individually looking perfectly valid.

I built Vault Steward to help detect those problems before they make a vault difficult to trust.

Vault Steward is an open-source, local-first Obsidian plugin that scans your vault for structural and knowledge-integrity issues, presents cited findings, and gives you safe repair options.

It never edits a note on its own.

What it checks

Vault Steward currently checks for:

  • broken internal links, embeds, and anchors

  • Markdown and frontmatter schema violations

  • task and decision drift

  • deterministic YAML policy issues

  • duplicate entities

  • contradictions

  • stale knowledge

  • ambiguous decisions

Some checks are deterministic.

For cases where semantic reasoning is useful, a configured model can classify or rank bounded, cited evidence.

The model cannot write notes, change policy, approve a proposal, or apply a fix.

Review first, edit second

The workflow is intentionally:

Check → Evidence → Preview → Approve

For proposed repairs, Vault Steward shows the exact Current → After change before anything is applied.

You choose which fixes to accept.

If a note changes after a preview was generated, Vault Steward preserves the newer edit and requires a fresh preview rather than overwriting it.

Local-first

Ollama is the primary local-first model path.

Vault Steward also supports llama.cpp-compatible providers using loopback-only endpoints.

Optional cloud providers are explicitly opt-in and only receive the bounded evidence required for the selected request.

The plugin’s canonical state remains local.

Why I built it

The problem I wanted to solve wasn’t simply having too many notes.

It was reaching the point where a vault contains enough accumulated history that you can no longer be certain everything inside still agrees.

A broken link is obvious.

Two individually valid notes that quietly disagree with each other are much harder to notice.

Vault Steward is an attempt to treat an Obsidian vault as a knowledge system that needs ongoing integrity checks, rather than a collection of files we assume will remain consistent forever.

I’d particularly like feedback from people with older or larger vaults.

What kinds of problems tend to accumulate in your vault that are difficult to catch manually?