Vault Therapist

AI-powered vault analyst for Obsidian. Find orphan notes, detect contradictions, surface knowledge gaps, and get a weekly insight report on your second brain.

Vault Therapist reads the notes in your vault and runs four AI-powered analyses designed for knowledge workers, researchers, and long-time note-takers whose vaults have grown faster than their memory of them.

---

Features

• Orphan finder — notes with no inbound or outbound links, plus AI-suggested connections to the rest of your vault.
• Contradiction detector — finds pairs of notes that make conflicting claims, with a confidence score and a plain-English explanation.
• Knowledge gap finder — scans clusters of notes around a topic and highlights what's missing or underdeveloped.
• Weekly insight report — themes, patterns, and recommendations across your vault, generated on demand.
• Four AI providers — Ollama (100% local), OpenAI, Anthropic, OpenRouter. Your notes never go anywhere you don't authorize.
• Embedding-powered similarity — once, then cached. Large vaults process incrementally.

Data privacy & network use

Disclosure (required by Obsidian's developer policy):

Vault Therapist makes outbound network requests to:

1. The AI provider you configure (Ollama/OpenAI/Anthropic/OpenRouter) — to run analyses. Your note content is sent to this provider. If you pick Ollama running on localhost, nothing leaves your machine.
2. api.vaulttherapist.com — only to validate your license key and report activation status. No note content is ever sent here.

A paid license is required for full access. A 3-day trial is included — no account, no credit card.

Read the full privacy policy: https://vaulttherapist.com/privacy

AI Disclaimer

Vault Therapist uses AI models to generate suggestions and analysis results. All AI output should be treated as suggestions, not authoritative conclusions. Always verify recommendations against your actual notes before making changes to your vault.

Install

Once accepted to the Obsidian community plugin directory:

1. Settings → Community plugins → Browse, search "Vault Therapist", install, enable.

Until then, or for beta access:

1. Download main.js, manifest.json, and styles.css from the latest release.
2. Drop them into <vault>/.obsidian/plugins/vault-therapist/.
3. Enable in Settings → Community plugins.

AI provider setup

Pick whichever matches your priorities (cost, privacy, model quality). You can switch at any time.

Ollama (local, free, fully private)

Best for privacy-first users. Slower on consumer hardware; cost is $0.

1. Install Ollama from https://ollama.com/download.

2. Pull a chat model and an embedding model:

   ollama pull gemma3:4b
   ollama pull nomic-embed-text
   

3. Make sure Ollama is running (ollama serve, or the tray app).

4. In Vault Therapist settings, pick Ollama, leave endpoint as http://localhost:11434.

5. Hit Test connection. Done.

OpenAI

Best quality/speed balance. Pay-as-you-go.

1. Create an API key at https://platform.openai.com/api-keys.
2. In settings, pick OpenAI, paste the key, set model (default gpt-4o).
3. Test connection.

Anthropic (Claude)

Strong reasoning; good for contradiction detection.

1. Create an API key at https://console.anthropic.com/settings/keys.
2. In settings, pick Anthropic, paste the key, set model (default claude-4-5-haiku).
3. Test connection.

OpenRouter

Single account, dozens of models (including free tiers).

1. Create an API key at https://openrouter.ai/keys.
2. In settings, pick OpenRouter, paste the key, set model (default google/gemma-4-31b-it:free).
3. Test connection.

Google Gemini

Google's in-house model family. Fast and cost-effective.

1. Get an API key from https://aistudio.google.com/app/apikey.
2. In settings, pick Google Gemini, paste the key, set model (default gemini-2.0-flash).
3. Test connection.

LMStudio (local)

Run LLMs locally on your machine with a polished app. No API key needed — just start the local server.

1. Download LM Studio from https://lmstudio.ai/download.
2. Open LM Studio and load a model (e.g. Llama 3.2, Mistral).
3. Click Start Server (the local API icon on the left sidebar).
4. In Vault Therapist settings, pick LMStudio, set Base URL to http://localhost:1234 (default), enter the model name exactly as shown in LM Studio.
5. Test connection.

Note: Embeddings depend on the loaded model — not all models support embeddings. The plugin will probe automatically on first use and disable embeddings if the model doesn't support them.

Pricing

	Trial	Lifetime license
Price	Free, no signup	$15 one-time
Duration	3 days	Forever, all future updates
Features	All	All
Vaults	Unlimited	Unlimited
Devices	This install	Activations across your devices

Buy: https://vaulttherapist.lemonsqueezy.com

FAQ

Does my note content leave my machine? Only if you pick a cloud AI provider. With Ollama, nothing ever leaves your machine. With OpenAI/Anthropic/OpenRouter, the notes being analyzed are sent to that provider under their terms. The Vault Therapist license server only ever sees your license key — never your notes.

Which provider should I pick?

• Privacy > everything: Ollama or LMStudio.
• Best general results for the price: OpenAI (gpt-4o).
• Best for subtle contradictions: Anthropic (claude-4-5-haiku).
• Want to try many models on one bill, including free tiers: OpenRouter.
• Fast, cheap, Google's model family: Google Gemini.
• Want to run everything locally with a GUI: LMStudio.

What happens when the trial expires? Analyses are gated, but the plugin keeps loading cleanly. Settings stay accessible so you can enter a license. No features crash, no data is lost, and you can uninstall without side effects.

What about refunds? 14-day no-questions-asked refund via Lemon Squeezy. Reply to the purchase email.

Why does the plugin call api.vaulttherapist.com? Only to verify your license key after activation and, occasionally, to confirm it's still valid. If the backend is unreachable, Vault Therapist falls back to a 24-hour offline grace window so a flaky connection never locks you out mid-session.

Is source code available? The plugin source is in this repo under the Business Source License 1.1 (BUSL-1.1). You can read, modify, and use it for non-production purposes; production use requires a commercial license. The code automatically converts to Apache 2.0 on the Change Date (2030-04-15), or four years after each version's release — whichever comes first. The license backend is proprietary.

Can I use it on mobile? Not yet — desktop only for now (isDesktopOnly: true). Mobile support is on the roadmap once embedding performance on-device is acceptable.

Troubleshooting

Embedding stalls or fails on large vaults (Ollama)

Ollama runs embedding models in memory. On large vaults (500+ notes), the embedding model may exhaust available RAM or GPU memory mid-run, causing Ollama to stop responding.

What you'll see: A notice like "Ollama stopped responding after embedding 996 of 1226 notes." The notes that were embedded before the failure are preserved — you don't lose progress.

How to fix it:

1. Restart Ollama and re-run. The plugin will skip notes that already have embeddings and continue from where it stopped.
2. Use a smaller embedding model. all-minilm (22 MB) uses far less memory than nomic-embed-text (274 MB) and still produces good similarity results:

   ollama pull all-minilm
   

   Then set all-minilm as your embedding model in Vault Therapist settings.
3. Close other applications to free memory, or re-run with fewer notes per batch.

The plugin automatically retries a failed batch once and falls back to embedding notes one-at-a-time if the batch still fails.

Which providers support native embeddings?

Provider	Native embeddings	Notes
Ollama	✅ Yes	Any model pulled via ollama pull <model>
OpenAI	✅ Yes	Uses text-embedding-3-small automatically
Google Gemini	✅ Yes	Uses text-embedding-004 automatically
LMStudio	⚠️ Varies	Only when the loaded model supports embeddings
OpenRouter	❌ No	Falls back to TF-IDF (approximate similarity)
Anthropic	❌ No	Falls back to TF-IDF (approximate similarity)

What does "TF-IDF fallback" mean?

When your AI provider doesn't offer a native embedding API (Anthropic, OpenRouter), Vault Therapist computes TF-IDF vectors instead. These use word-frequency statistics — they work for finding notes with overlapping keywords, but they're less accurate for semantic similarity than real neural embeddings. If you see a TF-IDF notice, consider switching to a provider with native embedding support for better results.

Recommended setup for very large vaults (1000+ notes)

If you have a large vault and Ollama keeps running out of memory, OpenAI's text-embedding-3-small is the most cost-effective option:

• Cost: ~$0.02 per 1M tokens — a 1000-note vault costs under $0.10 to embed.
• Speed: Batches process in seconds, not minutes.
• Quality: Better similarity results than TF-IDF or small local models.

Set your AI provider to OpenAI in Vault Therapist settings. Your chat model can still be any provider (Ollama, Anthropic, etc.) — only embeddings need OpenAI.

License

Business Source License 1.1 (BUSL-1.1). Converts to Apache License 2.0 on 2030-04-15 (or four years after each version's release, whichever comes first). Non-production use is free; production use requires a commercial license — see LICENSE for terms and contact support@vaulttherapist.com for licensing.

Support

• Bugs / feature requests: https://github.com/rzgrozt/vault-therapist/issues
• Email: support@vaulttherapist.com