Auto Tag¶
What This Feature Does¶
Auto Tag scans each finished transcription for configured keywords or phrases and attaches matching tags. It is deterministic organization, not model-based classification, and it never rewrites the transcript.
Tags are retained when the profile writes a persistent Note.
Quick Setup¶
- Open
Profilesfrom the status-bar menu and select the profile used for dictation or HTTP processing. - Enable
Save to Notes. - Choose the destination folder within the notes root and the artifacts to include.
- To add tags automatically, enable
Auto Tag, select an Auto Tag Library (or add inline fallback rules), and configure its phrase-to-tag rules. - Complete a dictation or send an HTTP transcription request using that profile.
- Open
Notes...from OpenScreech Home or the menu bar to browse the saved note.
Design Rationale¶
- Tags are deterministic phrase matches, not model output: fast, explainable, and stable, matching the word replacement philosophy.
- Tagging never modifies the transcription text; it only annotates the saved note.
- Note bundles are plain folders and files, so they remain portable, inspectable in Finder, and independent of a private database.
Pipeline Position¶
Auto Tag runs on the post-processed text:
word replacement -> selected final output -> auto tag -> route matching -> primary delivery
| | |
+-> early artifacts +-> final artifact +-> ordered routing actions (background)
Because it runs after LLM post-processing, rules match the cleaned-up text (including any LLM rewriting), not the raw ASR output.
Saved Outputs and Notes are not a final serial step after routing. Initial audio and transcript artifacts can be written earlier, and a Note can be updated with the selected LLM result and tags as they become available. Trigger action completion does not gate those writes.
Auto Tag Rule Fields¶
| Field | Meaning | Typical Use |
|---|---|---|
| Keyword or phrase | Text to look for in the transcription | standup |
| Tag to add | Tag attached to the saved note when the phrase appears | meetings |
| Enabled | Include/exclude rule | Temporary A/B testing |
| Case sensitive | Match exact casing only | Acronyms like OKR |
Matching is case- and diacritic-insensitive unless case sensitivity is enabled. Duplicate tags are collapsed (case-insensitively) while preserving rule order.
Rule Sources: Library vs Inline¶
- Each profile selects one shared Auto Tag Library (managed in
Auto Tag Libraries...). - Each profile can also hold inline fallback rules, used only when the selected library has no rules.
- Libraries import/export as JSON for backup and reuse.
How Tags Are Stored¶
Auto Tag runs whether or not Notes saving is enabled, but computed tags need a persistent destination to remain useful. Enable Save to Notes on the same profile to keep them.
See Notes for bundle contents, search, folders, transcript variants, audio, deletion behavior, and the distinction between Notes, temporary history, and flat-file Saved Outputs.
Common Pitfalls¶
| Pitfall | Outcome | Fix |
|---|---|---|
| Enabling auto tag without note saving | Tags are computed but have nowhere to persist | Enable Save to Notes on the same profile |
| Over-broad phrases | Notes accumulate noisy tags | Prefer specific multi-word phrases |
| Expecting tags to edit text | Tags never change the transcription | Use word replacement or LLM post-processing for text changes |
| Changing the notes root and seeing an empty list | Root changes do not move existing bundles | Switch back or move the bundles separately |
| Expecting file or meeting transcription results | Those flows do not currently write Notes bundles | Use dictation or the HTTP transcription server |
Validation Checklist¶
- A dictation containing a rule phrase produces a note with the expected tag in the Notes window.
tag:search narrows the list to matching notes.- Disabling
Save to Notesstops new notes while flat-file Saved Outputs continue unchanged.
Quick Tips¶
- Auto Tag evaluates the text after LLM post-processing. If a model rewrites a trigger phrase, write the rule for the final wording or move that classification into the model output.
- Tags are computed even when Notes is off, but they have no persistent destination. Enable
Save to Noteson the same profile when the label needs to survive.
See Also¶
- Capture and Organise Voice Notes
- Profiles
- Notes
- Word Replacement Libraries
- HTTP Transcription Service
- Data and Output Locations
Visual Notes¶
- TODO: add screenshot of the Notes window with tag search active.
- TODO: add screenshot of the Auto Tag section in a profile.