Skip to content

advanced-settings.md

Latest commit

 

History

History
154 lines (104 loc) · 5.06 KB

advanced-settings.md

File metadata and controls

154 lines (104 loc) · 5.06 KB
Raw

Advanced Settings

Settings available in VS Code Settings (Cmd/Ctrl+, → search Promptimize) that are not exposed in the configuration webview, plus details on planned settings and test output.

Transcription

promptimize.transcriptionLanguage
Default auto
Values auto, en, es, fr, de, it, pt, ja, ko, zh
Applied Yes — passed to OpenAI Whisper
Webview Not available — configure in Settings

When set to auto, Whisper auto-detects the spoken language.

{
  "promptimize.transcriptionLanguage": "en"
}

promptimize.transcriptionHint
Default (empty)
Applied Yes — passed to Whisper as the prompt parameter
Webview Not available — configure in Settings

Optional hint text to improve Whisper accuracy for domain-specific vocabulary, acronyms, or technical terms. Whisper uses this as context, not as content to transcribe.

Example: Project uses unusual terms or stack-specific jargon:

{
  "promptimize.transcriptionHint": "Kubernetes, Helm, ArgoCD, TypeScript, NestJS"
}

Tips:

  • List proper nouns, product names, and acronyms you expect to say
  • Keep hints concise (Whisper has a prompt length limit)
  • Does not replace transcriptionLanguage — use both when needed

Transformation

promptimize.transformationSystemPrompt
Default Built-in prompt engineer template
Applied Yes — sent to all optimization providers
Webview Editable in configuration panel

Customize transformation style, structure, and rules. See Configuration Webview Guide.

Context-Aware Optimization

When using Promptimize mode, the extension passes editor context to the optimization provider:

Context Source Purpose
editorLanguage Active editor's document.languageId Tailor prompt terminology (e.g. typescript, python)
projectType Derived from language Additional framing for the transformer

Requirements:

  • An editor tab must be active when you stop recording
  • If no editor is open, optimization still runs without language context

This is automatic — no setting required.

Dynamic Model Loading

Some providers fetch model lists at configuration time:

Provider Model list source
OpenAI Live API (listGptModels)
Ollama GET /api/tags on configured base URL
OpenCode GET /v1/models on proxy base URL
OpenRouter OpenRouter models API
Anthropic Static curated list
Google Static curated list
Azure Deployment name from settings
Cursor Static curated list

Use Refresh models in the configuration webview or re-run provider configuration commands after adding new local models.

Planned Settings (Not Yet Applied)

These settings appear in VS Code Settings and are loaded by the extension, but do not change runtime behavior yet:

Setting Default Planned behavior
promptimize.audioQuality high Adjust recording sample rate / encoding
promptimize.maxRecordingDuration 120 Auto-stop recording after N seconds
promptimize.showNotifications true Suppress progress toasts when false

Current behavior:

  • Audio is always captured at 16 kHz mono (optimal for Whisper)
  • Recording stops only when you click stop or cancel
  • Progress notifications always appear during stop/processing

Configure these in Settings for forward compatibility; they will take effect in a future release.

Test Configuration Output

Promptimize: Test Configuration validates Whisper and optimization, then opens a Configuration Test Result webview when optimization succeeds.

Webview contents

Section Description
Status lines ✓ Whisper (OpenAI): Working and ✓ Optimization (Provider): Working
Cost note Reminder that Whisper charges apply; optimization test uses sample text only
Original Sample developer ramble sent to the transformer
Transformed Optimized prompt from your configured provider
Improvements Heuristic list of detected enhancements (see below)

Improvements heuristics

The improvements list is computed locally (not from the LLM). Possible entries:

Improvement Trigger
Removed filler words Original contained fillers (um, uh, like, etc.) and transformed text has fewer
Made more concise Transformed text is at least 10% shorter
Added clear structure Transformed text includes Context:, Objective:, or Requirements: sections
Improved sentence structure Transformed text has more sentences than the original

If optimization is disabled, the command tests Whisper only and shows an information toast.

See also: Configuration Guide · Configuration Webview Guide