Wispr Flow custom dictionary: a developer setup guide

If Flow keeps misspelling a framework, API or colleague's name, first check whether the error repeats the same way. Adding the intended word and replacing a recurring wrong spelling are two different dictionary actions.

This guide uses Wispr documentation checked on September 11, 2026. The downloadable vocabulary is an editorial starter list, not a tested configuration or a claim that Flow misrecognizes these words. Our Wispr Flow review covers the broader limits of dictation and text cleanup.

Choose between vocabulary and a correction rule

According to Wispr's dictionary documentation, vocabulary entries help recognition of uncommon words and phrases. The separate Correct a misspelling option replaces a specified wrong spelling after dictation.

Start with a vocabulary entry when the term is unusual or errors vary. If the output repeatedly contains the same wrong spelling, consider a correction rule. For example, you might add Supabase as vocabulary. Only set a rule replacing “super base” if that is an error you have actually observed; this is a hypothetical example, not a Flow test result.

A replacement also needs a counterexample. If the supposedly wrong phrase is valid in another context, read a sentence where you want it preserved. An aggressive rule can fix a product name while changing ordinary prose incorrectly.

Add a term on desktop or mobile

On Mac and Windows, open Flow → Dictionary → Add new, enter the word or short phrase and click Save. Use your personal dictionary while trying entries. The documentation distinguishes personal entries from shared team entries and says personal entries take precedence for the same word.

On iOS, open Dictionary at the bottom of the app, tap +, enter the term and tap Save. On Android, open Dictionary, wait for the list to load, tap +, enter the term and tap Add. Android rejects duplicate entries and words already used as snippet triggers, among other validation errors.

Dictate a complete sentence containing the term after saving. Testing an isolated word is less representative if you normally use it inside a coding prompt. Wispr says dictionary entries sync across devices, but verify that an entry appears on the second device before testing there. A word added during an Android session applies from the next session.

Add a rule only after recording the wrong output

On desktop, choose Dictionary → Add new, enter the correct spelling, enable Correct a misspelling, enter the wrong spelling and save. For an existing or imported desktop entry, open it and add the correction there. Wispr documents one replacement rule per word, so edit the existing entry when changing that correction.

On iOS, the documentation says the correction toggle appears when adding a new word. Android applies synced rules but does not create them; create the rule on desktop or iOS. Do not assume identical editing controls across platforms.

Keep the original sentence and raw output in a note before changing anything. After the rule is saved, repeat the sentence and the counterexample. If a correction creates new mistakes, edit or remove it and retest with vocabulary alone.

Prepare a small developer CSV

The 20-word developer vocabulary file includes terms such as API, APIs, TypeScript, PostgreSQL, Supabase, webhook and n8n. It contains one term per row, with no header. Remove terms you do not use before importing; twenty entries are a starting set, not a recommended minimum.

Keep the singular and plural forms you actually need. Wispr specifically recommends adding both forms for terms such as API and APIs. For exact casing, paths and symbols, inspect the output rather than assuming a dictionary entry guarantees code-safe text.

The current desktop instructions are:

1. Go to Settings → Experimental and enable Bulk Import.
2. Return to Dictionary, choose the personal destination and select Import.
3. Choose the CSV or drag it into the upload area.
4. Review the preview and destination before completing the import.
5. Check a few imported words in the list, then dictate sentences containing them.

Wispr documents one-column word files and two-column misspelling/correction files. It skips duplicates and malformed rows; the preview's duplicate count is not a complete report of malformed input. The supplied file intentionally uses vocabulary only, so it does not install guessed replacement rules. Its CSV structure has been checked locally; import behavior has not been independently tested in the app for this article.

Bulk import is documented for Mac and Windows, not iOS or Android. If the option is absent from your desktop version, add the small set manually and consult the current help page rather than following a mobile import tutorial.

Compare the same sentences before and after

Use the blank dictation results log to record the app version, microphone, language selection, exact sentence and unedited output. Mark the conditions dictionary_before and dictionary_after, and record which entries and rules changed.

Choose five terms you use often. Read each inside a normal sentence before adding it, add only those entries, then read the same sentences again. Repeat if practical. Keep microphone, target app and language selection unchanged so you are not testing several configuration changes at once.

Score whether each required term survives exactly and how long the entire sentence takes to correct. Include a sentence that does not contain your target terms and any counterexamples to replacement rules. Blank measurement fields mean “not measured,” not zero. This procedure produces evidence for your configuration; it does not establish universal recognition accuracy.

For Vietnamese prompts containing English terms, keep language selection as a separate experiment using the Vietnamese and mixed-language setup guide. Changing languages and dictionary entries simultaneously makes an improvement harder to explain.

Know when the dictionary is the wrong fix

A vocabulary entry cannot restore audio that was never recorded or make a cloud transcription work offline. It also cannot guarantee that automatic text cleanup preserves a negation, date or instruction. Those limits are covered in Wispr's accuracy documentation.

If recognition is the problem, upgrading should not be your first experiment: Wispr's plan documentation says all tiers use the same recognition engine. Our free versus Pro guide explains what a paid tier changes.

Start with the next word that actually fails in your work. Record it, add the smallest useful entry or rule, and keep the change only if the full sentence requires less correction without introducing new errors.

For IDE context and file references, see the Cursor dictation workflow. If nothing is inserted at all, diagnose recording, transcription and paste failures before changing vocabulary.