Dictating prose is easy. Dictating handleTokenRefresh is not — a general
speech recogniser hears three ordinary words and writes three ordinary words.
Vibe Coding closes that gap. It looks at the identifiers already visible in your editor and tells the recogniser what to expect, so the name you say comes back as the name that exists in your code.
It is on by default and lives in Settings → Vibe Coding.
What Vibe Coding changes
When Code Context Awareness is on, Sprecho captures three things from the window you are dictating into:
- Function and variable names visible on screen
- The naming convention in use — camelCase, snake_case
- The programming language, from the file extension
Those symbols are handed to the speech recogniser as expected vocabulary before your dictation is processed. The effect is the same as adding every identifier in the open file to your dictionary, except it happens automatically and changes as you move between files.
It does not rewrite your code, generate code, or turn dictation into commands. It makes transcription accurate in a context where accuracy is otherwise poor.
Which editors and terminals
Vibe Coding works with code editors and terminals — VS Code, Cursor, Xcode and Terminal are the ones Sprecho classifies as the Code category out of the box, and the same category drives which formatting model runs there.
For the Code category, Standard is the right model: it cleans up punctuation and layout without rewriting your wording, which is exactly what you want when half of what you dictate is an identifier.
Vibe Coding helps the recogniser, not the formatter. If your prose comes out over-polished in an editor, that is the model setting on the Models page, not this toggle.
Excluding directories
Sprecho skips a list of directories when analysing code context. The defaults are the ones that would otherwise flood the vocabulary with noise:
node_modules · .git · dist · build · __pycache__ · .venv
To add your own, open Settings → Vibe Coding, type the path under Excluded Directories and press Enter or click the plus button. Click the × on a chip to remove it.
Add anything vendored, generated or minified. Symbols from a bundled library compete with the ones you actually dictate.
Privacy
This is the part developers ask about first, so it is worth stating plainly:
- All code analysis happens locally on your device.
- No actual source code is ever sent to any server.
- Only extracted symbol names are used, and only to improve transcription accuracy.
Speech audio and the resulting text still go to Sprecho's servers in Germany for recognition and formatting, exactly as with any other dictation — see What Sprecho stores and where.
Turning it off, and admin locks
Open Settings → Vibe Coding and switch off Code Context Awareness.
In a managed organisation, an administrator can enforce this setting centrally. When that is the case the toggle is greyed out with a note that it is managed by your team admin, and flipping it locally has no effect — the policy is re-applied. See The Enterprise admin dashboard.
Still on the roadmap
Two items are visible in the settings page under Coming Soon and are not available yet:
- Project indexing — indexing a whole project for context across files.
- Editor extensions — VS Code and JetBrains extensions for deeper integration.
Today's Vibe Coding works from what is on screen, which needs no extension and no indexing step.