Nathan Cheng951 downloadsMatch and apply tags to one note or many with AI decision models (default: Jev), using your vault’s tags or a specific set you provide. Faster + cheaper than LLM-based taggers.
Tag Match finds relevant tags for notes in your Obsidian vault. You can use existing vault tags or provide a specific set, including tags you haven't used yet.
You can review suggestions for an individual note or add matches directly to one note or a batch.
Generating tags note by note can lead to overlapping labels and a bloated tag vocabulary. Tag Match helps you maintain a consistent vocabulary by working from only existing or provided tags, leaving tag creation to you. The included assistant prompt helps you review their meanings and refine your tagging rules as your system evolves.
You need Obsidian 1.13.0 or later and an account with TypeSafe, OpenRouter, OpenAI, or Gemini. Analysis requires an API key and paid credits with your chosen provider. Jev via TypeSafe is the default model.
To review tags for your first note:

To add matches without reviewing them, run Tag Match: Match and add tags to current note. Both commands use your tag selection settings and refuse to apply results if the note has changed.
With Skip short notes on (the default), adding tags without review skips notes whose description and body together have fewer than 20 words, because their tags would rest mostly on the title. Analyze those notes to review their suggestions. Change the cutoff in Minimum words, or turn Skip short notes off.
To apply tags to a batch:
Use Run in background to keep working. Reopen progress through the notice, desktop status bar, or Tag Match: Show bulk tagging progress command.
Stop cancels remaining work. Requests already running may still use credits. Results show additions, skipped notes, and failures. Notes under the minimum word count are skipped without a request.
Undo additions restores notes that haven't changed since tagging. Undo remains available until you close completed results, reload the plugin, or quit Obsidian.
Tag Match uses decision models to score each tag without generating prose. Under Settings → Tag Match → Connection and agents, each provider has its own API key and model fields. Switching providers preserves the other providers' settings.
To change models, enter a compatible model ID in the provider's model field. The link beside the field lists supported models.
| Connection provider | Default model | Compatible model IDs |
|---|---|---|
| TypeSafe | jev-latest |
TypeSafe models and aliases |
| OpenRouter | ~typesafe/jev-latest |
OpenRouter Jev latest alias |
| OpenAI | gpt-6-luna |
OpenAI Decisions API models |
| Gemini | gemini-decision-1 |
Gemini Decision model |
OpenRouter routes requests to a hosting provider. Model availability also depends on your OpenRouter account's provider and privacy settings. General chat models are not supported by Tag Match's decision requests.
Model limits and response times differ. Large tag selections use multiple requests, processed in sequence for each note. Tag Match retries with smaller requests when it recognizes a provider's tag-count limit error. If analysis fails, Tag Match returns no partial suggestions for that note.
Selection method sets how many vault tags to consider. Choose the default, a specific number or percentage, or All tags. Considering more tags can find more matches, but can take longer and use more API credits.
When the selection is limited, Selection mix lets you balance your most-used tags with a sample of the rest. The sample varies between notes, giving less-used tags a chance without checking your entire vocabulary every time.
After a review, Score more tags draws another random sample of the same size from tags that haven't been scored yet.

Only these tags in Selection method uses the tags you supply, including tags not yet used in the vault. It replaces vault selection and sampling, and ignores Excluded tags. Tags already on the note are skipped with either selection method.
For vault tag selection, Excluded tags removes tags before analysis. For example, admin excludes that tag, and work/* excludes work and its descendants. Tagging guidance influences the model's judgment, while exclusions prevent the plugin from considering those tags.
Adjust for this review and Adjust for this batch override your saved settings for one run.
Tag Match chooses candidate tags locally, then asks the decision model to estimate the probability that each tag belongs on the note. The model receives the note's title, description, sampled body, and existing tags. It evaluates each candidate tag using your general guidance, any definition you supplied for that tag, and criteria for a meaningful match. Because tags are evaluated independently, several can qualify for the same note.
Tag Match ranks those probabilities. Minimum match score sets how strong a match must be, while Maximum tags to add limits how many are preselected for review or added directly. You can change the selection in review mode before adding tags.
Use short rules for general guidance, such as "Tag substantial topics. Skip passing mentions." Give individual tags a definition when your use differs from their ordinary meaning:
dev = Building, debugging, or maintaining software. Include implementation tutorials. Exclude general technology news without development content.

Long notes are sampled from the beginning, middle, and end within your text limit. The review shows how many tags will be considered and whether the note was sampled before analysis.
Agents can review or apply tags to one note or a batch through the optional Node CLI. It uses the plugin's saved provider, model, tagging preferences, and Obsidian Secret reference. The CLI reads the Secret while that vault is open in Obsidian. An environment variable can supply the key when Obsidian is closed.
To set up the CLI on desktop:
Obsidian updates the plugin separately from the CLI. Tag Match offers a CLI update when the installed companion does not match the plugin.
The agent guide covers commands, bulk tagging and undo, credentials, and use without Obsidian.
To install without the community plugin browser:
main.js, manifest.json, and styles.css from the latest release.<vault>/.obsidian/plugins/tag-match/.Analysis sends the note title, description, existing tags, selected body text, candidate tags, tagging guidance, and relevant tag definitions to your chosen service: api.typesafe.ai, openrouter.ai, api.openai.com, or generativelanguage.googleapis.com. With OpenRouter, requests are routed to the model provider. Analysis starts only when you invoke it. There is no background vault scan sent to a model.
API keys stay in Obsidian Secrets on the device where you enter them. The plugin's data.json stores only the selected secret's name. Synced settings may carry that name to another device, but not the key. Tag Match has no client-side telemetry. Provider data handling follows TypeSafe's privacy policy and, when selected, OpenRouter's privacy policy, OpenAI's API data policies, or the Gemini API terms.
Copy buttons write to the system clipboard only when clicked. Tag Match never reads it.
The Obsidian plugin operates within your vault. The optional CLI reads its explicit configuration path and can read or modify a Markdown file outside a vault when you supply that path. It does not search your filesystem for notes or credentials.
Tag Match is an independent project by Nathan Cheng. It is not affiliated with Obsidian, TypeSafe, OpenRouter, OpenAI, or Google.
Report a bug · Security policy · Third-party notices · Buy me a coffee