The Tools Tab
The Tools tab is the core workbench for prompt engineering. It contains four main views:
- Prompt Library — Browse and search hundreds of built-in and custom prompts organized by category.
- Prompt Template — Edit and run templates interactively using Mustache syntax. Fill in variables and see the completed prompt and response in real time.
- Prompt Scripting — Define a batch of prompt scripts and execute them in sequence against one or more models.
- Prompt Validator — Automatically validate model responses against expected outputs or custom criteria.
Prompt Templates with Mustache
Prompts are stored as Mustache templates, making it easy to define reusable prompts with variable placeholders:
Translate the following {{language}} text to English:
{{text}}In the Prompt Template view, fill in the variable fields on the right panel and click Run to send the completed prompt to the selected model.
Multi-Step Pipelines
The Prompt Scripting view lets you define a sequence of prompt steps that feed outputs from one step as inputs to the next. This is useful for:
- Summarizing a document, then extracting key entities from the summary
- Translating text, then assessing translation quality
- Generating a draft, then requesting a critique and revision
API Playground
Under the API tab, PromptFx exposes raw API endpoints for:
- Completions — Classic text completions
- Chat — Multi-turn chat completions with system prompts
- Embeddings — Compute and inspect embedding vectors
- Images — Generate images with DALL-E or Gemini Imagen
- Audio — Speech-to-text (Whisper) and text-to-speech
- Models — Browse available models and their capabilities
These views are similar to the OpenAI Playground and are great for exploring model behavior and comparing responses.
NLP Task Views
The Text tab provides high-level views for common NLP tasks:
- Translation (multiple target languages)
- Summarization (short, medium, and bullet-point formats)
- Sentiment analysis
- Named entity extraction
- Keyword extraction
- Classification
Configuring Models
Use the toolbar dropdowns at the top of the application to select the default completion model and embedding model. Most views respect these defaults, though each view's configuration panel (right side) also allows per-view overrides.
PromptFx supports OpenAI models (GPT-4, GPT-3.5, etc.) and Google Gemini models out of the box. To connect to a custom or locally hosted API, configure the base URL in the settings panel.