Prompts

Version-controlled AI prompt templates with variable substitution, input validation, and execution tracking.

A Prompt in Prompt Forge is a reusable, version-controlled template that defines how to interact with Large Language Models (LLMs). Each prompt consists of:

• Template Content - The actual prompt text with variable placeholders
• Input Schema - JSON Schema defining required and optional variables
• Model Configuration - Provider, model, temperature, max tokens, etc.
• Version History - Track changes over time with incremental versions
• Execution Metrics - Latency, token usage, and cost tracking

Basic Prompt Anatomy

You are a sentiment analysis expert. Analyze the sentiment of the following text.

Text: {{text}}

Provide your analysis in the following format:
- Sentiment: (Positive, Negative, or Neutral)
- Confidence: (0.0 to 1.0)
- Key Phrases: List the phrases that influenced your decision

Variables are denoted using double curly braces: {{variableName}}. These will be replaced with actual values when the prompt is executed.

Input Schema

The input schema defines what variables are required and their types. It follows JSON Schema specification:

{
  "type": "object",
  "properties": {
    "text": {
      "type": "string",
      "description": "The text to analyze for sentiment"
    }
  },
  "required": ["text"]
}

Model Parameters

Each prompt version includes configuration for the AI model:

{
  "model": "claude-3-5-sonnet-20241022",
  "temperature": 0.3,
  "max_tokens": 500
}

Via Web Interface

Navigate to Prompts → New Prompt in the dashboard:

1. Enter a unique name (will be converted to kebab-case)
2. Add a description of what the prompt does
3. Select a category (Summarization, Analysis, Generation, etc.)
4. Write your template with {{variables}}
5. Define input schema for variables
6. Configure model parameters
7. Test with sample inputs
8. Publish when ready

Via GraphQL API

# Step 1: Create the prompt
mutation CreatePrompt {
  createPrompt(input: {
    name: "sentiment-analyzer"
    description: "Analyzes sentiment of text with confidence scores"
    category: ANALYSIS
    isPublic: false
  }) {
    id
    name
  }
}

# Step 2: Add first version with content
mutation CreatePromptVersion {
  createPromptVersion(input: {
    promptId: "cm1234567890"
    content: """
You are a sentiment analysis expert. Analyze the sentiment of: {{text}}

Format:
- Sentiment: (Positive/Negative/Neutral)
- Confidence: (0.0-1.0)
"""
    params: {
      model: "claude-3-5-sonnet-20241022"
      temperature: 0.3
      max_tokens: 300
    }
    inputSchema: {
      type: "object"
      properties: {
        text: {
          type: "string"
          description: "Text to analyze"
        }
      }
      required: ["text"]
    }
  }) {
    id
    version
  }
}

Prompts support versioning to track changes over time. Each version is immutable once created. This allows you to:

• Safely iterate on prompts without breaking existing integrations
• Roll back to previous versions if needed
• A/B test different prompt variations
• Track performance improvements across versions

Creating New Versions

When you update a prompt through the web interface, you're creating a new version. The version number automatically increments:

v1 → Initial version
v2 → Improved clarity of instructions
v3 → Added examples for better consistency
v4 → Optimized for lower token usage

Execute a prompt by providing values for all required variables:

mutation ExecutePrompt {
  executePrompt(
    promptId: "cm1234567890"
    input: {
      text: "I absolutely love this product! Best purchase ever!"
    }
  ) {
    output
    latencyMs
    tokenIn
    tokenOut
    costUsd
    executionId
  }
}

Response Structure

{
  "data": {
    "executePrompt": {
      "output": "- Sentiment: Positive\n- Confidence: 0.95",
      "latencyMs": 1250,
      "tokenIn": 45,
      "tokenOut": 28,
      "costUsd": "0.0012",
      "executionId": "exec_abc123"
    }
  }
}

Prompts can be organized into categories for better organization:

You can fork any prompt (including public prompts from other users) to create your own copy:

mutation ForkPrompt {
  forkPrompt(id: "cm1234567890") {
    id
    name
    versions {
      id
      version
    }
  }
}

Forking creates a complete copy with all versions, which you can then modify independently.