Skip to content
ALTAIRA LABS
Docs Blog
PromptPack PromptPack PromptPack

File Format & Portability

PromptPacks use JSON as their file format, optimized for both human readability and machine processing. This choice isn’t arbitrary—JSON provides specific benefits that make PromptPacks practical for real-world AI development and deployment.

Why JSON?

Universal Compatibility

JSON works everywhere:

  • Every programming language has robust JSON support
  • All cloud platforms can process JSON natively
  • CI/CD pipelines can parse and validate JSON files
  • APIs and databases accept JSON without conversion
  • Developer tools provide excellent JSON editing support

Machine and Human Friendly

PromptPacks need to be both executable by systems and readable by developers. JSON strikes this balance perfectly:

{
  "id": "customer-support",
  "name": "Customer Support Pack",
  "version": "1.0.0",
  "prompts": {
    "support": {
      "system_template": "You are a helpful {{role}} for {{company}}.",
      "variables": [
        {
          "name": "role",
          "type": "string",
          "required": true,
          "description": "The assistant's role"
        }
      ]
    }
  }
}

Human benefits: Clean structure, clear hierarchy, easy to read and edit Machine benefits: Fast parsing, strong validation, efficient processing

Validation and Tooling

JSON’s mature ecosystem provides established tooling:

  • Schema Validation: Catch errors before deployment using JSON Schema
  • IDE Support: Syntax highlighting, auto-completion, real-time validation
  • Command-line Tools: jq, validation utilities, formatters
  • Version Control: Clean diffs, merge conflict resolution

File Structure Benefits

Single-File Deployment

Everything your AI system needs is in one file:

{
  "prompts": { "support": {...}, "sales": {...} },
  "tools": { "lookup_order": {...}, "create_ticket": {...} },
  "fragments": { "greeting": "...", "policies": "..." },
  "template_engine": { "version": "v1", "syntax": "{{var}}" }
}

Deployment advantage: Copy one file, deploy complete AI system. No missing dependencies, no broken references, no configuration drift.

Version Control Friendly

JSON diffs clearly show what changed:

{
  "prompts": {
    "support": {
      "system_template": "You are a support agent.",
      "system_template": "You are a friendly support agent.",
      "parameters": {
        "temperature": 0.7
        "temperature": 0.8
      }
    }
  }
}

Development benefits: Clear change history, easy code reviews, confident rollbacks

Practical Format Guidelines

File Naming

Use descriptive names that indicate purpose and version:

customer-support-v1.pack.json      # Recommended
support-pack-2025-10-31.json       # Date-based versioning
cs-pack.json                       # Too generic
mypack.json                        # Not descriptive

Internal Organization

Structure your JSON for maintainability:

{
  // Identity first - what this pack does
  "id": "customer-support",
  "name": "Customer Support Pack",
  "version": "1.0.0",
  "description": "Complete customer service solution",


  // Shared configuration
  "template_engine": { ... },


  // Core content - prompts grouped logically
  "prompts": {
    "support": { ... },    // General support
    "technical": { ... },  // Technical issues
    "billing": { ... }     // Payment problems
  },


  // Supporting resources
  "tools": { ... },
  "fragments": { ... }
}

Readability Practices

Use consistent formatting:

  • 2-space indentation for readability
  • Logical ordering of fields
  • Descriptive property names
  • Clear variable naming

Add context through structure:

{
  "prompts": {
    "support": {
      "name": "General Support Assistant",
      "description": "Handles general customer inquiries and directs to specialists",
      "system_template": "..."
    }
  }
}

Validation and Quality Assurance

Schema Validation

PromptPacks validate against JSON Schema, catching errors early:

Terminal window
# Validate before deployment
ajv validate -s promptpack.schema.json -d my-pack.json


# Validate in CI/CD pipeline
npm test -- --validate-packs

Catches:

  • Missing required fields
  • Invalid data types
  • Incorrect version formats
  • Malformed tool definitions
  • Variable validation errors

Development Workflow

Terminal window
# 1. Edit pack file
vim customer-support.pack.json


# 2. Validate structure
promptpack validate customer-support.pack.json


# 3. Test functionality
promptpack test customer-support.pack.json


# 4. Deploy
promptpack deploy customer-support.pack.json

Portability Benefits

Cross-Platform Deployment

The same pack file works across:

{
  "compilation": {
    "compiled_with": "packc-v1.0.0",
    "created_at": "2025-10-31T12:00:00Z",
    "schema": "v1"
  }
}
  • Local development (laptop, Docker)
  • Cloud platforms (AWS Lambda, Azure Functions, GCP Cloud Run)
  • Edge devices (if supported by runtime)
  • Different AI providers (OpenAI, Anthropic, local models)

Environment Configuration

Packs adapt to different environments through variable substitution:

{
  "prompts": {
    "support": {
      "system_template": "You work for {{company_name}} in {{environment}}.",
      "variables": [
        {
          "name": "company_name",
          "type": "string",
          "required": true
        },
        {
          "name": "environment",
          "type": "string",
          "default": "production"
        }
      ]
    }
  }
}

Runtime flexibility: Same pack, different configurations per environment.

Integration Patterns

API Integration

PromptPacks work naturally with REST APIs:

// Load and execute pack
const pack = await fetch('/api/packs/customer-support')
const prompt = pack.prompts.support
const response = await ai.complete(prompt, variables)

Database Storage

Store packs in databases for dynamic loading:

CREATE TABLE promptpacks (
  id VARCHAR PRIMARY KEY,
  version VARCHAR,
  pack_data JSONB,  -- Full pack as JSON
  created_at TIMESTAMP
);

Configuration Management

Integrate with configuration systems:

# Kubernetes ConfigMap
apiVersion: v1
kind: ConfigMap
metadata:
  name: ai-packs
data:
  customer-support.json: |
    {
      "id": "customer-support",
      "version": "1.0.0",
      ...
    }

Performance Considerations

File Size Optimization

  • Minimize whitespace in production files
  • Use fragments to avoid duplication
  • Split large packs if they exceed reasonable size (~1MB)

Loading Efficiency

  • Cache parsed packs in memory
  • Validate once at load time, not per request
  • Pre-compile templates if your runtime supports it

Network Transfer

{
  "metadata": {
    "compression": "gzip",
    "content_type": "application/json",
    "size_kb": 245
  }
}

Packs compress well due to JSON’s structured nature and repeated patterns.

Migration and Compatibility

From Other Formats

Converting existing prompt systems:

Terminal window
# From YAML
yq eval -o=json promptpack.yaml > promptpack.json


# From CSV/spreadsheets
python convert_prompts.py --input prompts.csv --output pack.json


# From plain text files
promptpack import --directory ./prompts/ --output pack.json

Version Evolution

JSON structure supports backward-compatible evolution:

{
  "version": "2.0.0",
  "compatibility": {
    "min_runtime": "1.5.0",
    "deprecated_fields": ["old_field"],
    "migration_notes": "Use new_field instead of old_field"
  }
}

The JSON format makes PromptPacks a practical, production-ready solution for packaging and deploying conversational AI systems at scale.