Skip to main content

​
Prompt Management

MutagenT provides a complete prompt management system with versioning, variables, and lifecycle management. Centralize your prompts, track changes, and optimize performance systematically.

​
Prompt Lifecycle

Each stage in the lifecycle builds on the previous. Start with a draft, test in the playground, validate with datasets, optimize automatically, and deploy with confidence.

​
Key Features

Version Control

Track changes with full version history. Compare versions, rollback when needed, and maintain audit trails.

Template Variables

Dynamic content with type-safe variables. Create reusable templates that adapt to different inputs.

Evaluation

Measure quality with automated testing. Use G-Eval, semantic similarity, and custom metrics.

Optimization

Automatically improve prompts through AI-driven mutation and selection cycles.

​
Prompt Content Types

MutagenT supports three content types when creating a prompt. Use exactly one:
Content TypeFieldsUse Case
Raw PromptrawPromptSingle prompt text with {variables}
System + HumansystemPrompt + humanPromptStructured system instructions paired with a user-facing template
Messages ArrayArray of {role, content} objectsFull chat-format messages (system, user, assistant)

​
Prompt Structure

interface Prompt {
  id: number;                                // Unique identifier
  name: string;                              // Display name
  description?: string;                      // Optional description
  version: string;                           // Semantic version (e.g., "1.0.0")
  isLatest: boolean;                         // Whether this is the latest version
  systemPrompt?: string;                     // System-level instructions
  humanPrompt?: string;                      // User-facing prompt template
  rawPrompt?: string;                        // Raw prompt text
  inputSchema: Record<string, any>;          // JSON Schema for input variables
  outputSchema?: Record<string, any>;        // JSON Schema for expected output
  metadata?: Record<string, any>;            // Arbitrary metadata
  tags?: string[];                           // Organization tags
  createdBy?: string;                        // Creator email
  createdAt: string;                         // ISO 8601 timestamp
  updatedAt: string;                         // ISO 8601 timestamp
}

​
Quick Example

import { Mutagent } from '@mutagent/sdk';

const client = new Mutagent({ apiKey: process.env.MUTAGENT_API_KEY });

// Create a prompt with rawPrompt content type
const prompt = await client.prompt.createPrompt({
  name: 'Customer Support',
  rawPrompt: `You are a support agent for {company_name}.

Help the customer with their question:
{customer_question}

Be helpful, professional, and concise.`,
  inputSchema: {
    company_name: { type: 'string' },
    customer_question: { type: 'string' },
  },
  description: 'Main customer support prompt template',
});

console.log('Created prompt:', prompt.id);
The CLI warns if you accidentally use double-brace {{variable}} syntax. MutagenT uses single-brace {variable} syntax for template variables.

​
Prompt Organization

Organize prompts effectively within your workspace:
Use clear, descriptive names that indicate the prompt’s purpose:
  • customer-support-v1
  • code-review-assistant
  • data-extraction-json
Add descriptions to document:
  • What the prompt does
  • Expected inputs and outputs
  • Any special considerations
Define inputSchema and outputSchema as JSON Schema objects to document expected parameters and responses:
{
  "inputSchema": {
    "customer_name": {"type": "string"},
    "order_id": {"type": "string"}
  },
  "outputSchema": {
    "action": {"type": "string", "enum": ["refund", "replace", "escalate"]},
    "response": {"type": "string"}
  }
}

​
What’s Next?

Versioning

Learn how to track and manage prompt versions

Variables

Create dynamic prompts with template variables