PatBat - Constrained Code Generation

Tool for generating constrained code with TypeScript schema validation.

Installation

You can install the package using npm or bun:

# Using npm
npm install patbat-codegen

# Using bun
bun add patbat-codegen

CLI Usage

The CLI can be run directly using bun:

bun x patbat-codegen -i input.json -o output.json

CLI Options

• -i, --input <path> - Input JSON file path (required)
• -o, --output <path> - Output JSON file path (optional, defaults to <input>_output.json)
• -c, --config <path> - Config JSON file path (optional)
• -m, --model <model> - Override model specified in input file (optional)
• --no-validate - Skip input validation (optional)
• --return-code - Include the generated code in output (optional)

Package Usage

TypeScript/JavaScript

import { generateCode } from "patbat-codegen";

// Basic usage
const result = await generateCode({
  question: "What is the average age?",
  model: "claude-3-5-haiku-20241022",
  sampling: {
    maxTokens: 30,
    minPerFile: 1
  },
  input: {
    schema: "z.object({ name: z.string(), age: z.number(), city: z.string() })",
    data: [
      { "name": "Alice", "age": 25, "city": "New York" },
      { "name": "Bob", "age": 30, "city": "San Francisco" }
    ]
  },
  output: {
    schema: "z.object({ averageAge: z.string(), oldestPerson: z.string() })"
  }
});

// With config options
const resultWithCode = await generateCode({
  // ... same input as above
}, {
  validateInput: true,  // Validate input data against schema (default: true)
  returnCode: true      // Include generated code in output (default: false)
});

console.log(result);

Input Schema

type CodegenInput = {
  question: string;
  model: ModelTypeEnum;    
  apiKey?: string;         
  
  sampling?: {
    maxTokens: number;    
    minPerFile: number;    
  };

  input: {
    schema: string;        // Zod schema as string
    data: any[];          // Array of data objects
  };

  output: {
    schema: string;       // Expected output Zod schema
  };
}

// Available models
type ModelTypeEnum = 
  | "claude-3-haiku-20240307"
  | "claude-3-sonnet-20240229"
  | "claude-3-opus-20240229"
  | "claude-3-5-sonnet-20240620"
  | "claude-3-5-sonnet-20241022"
  | "claude-3-5-haiku-20241022";

Output Schema

type CodegenOutput = {
  requestId: string;
  requestedAt: string;
  output: any;             // Validated output matching output schema
  code?: string;           // Generated code when returnCode=true
  llmUsage: {
    input_tokens: number;
    output_tokens: number;
  };
  performanceTimeStamps: {
    startLLMCallPrep: number;
    endLLMCallPrep: number;
    startLLMCall: number;
    endLLMCall: number;
    startCodeRun: number;
    endCodeRun: number;
    responseReady: number;
  };
  model: ModelTypeEnum;
}

Development

Clone the repo and install dependencies:

git clone <repository-url>
cd patbat-codegen
bun install

Available Scripts

• bun run build - Build the package
• bun run cli - Run the CLI directly from source

License

Private use license - see LICENSE.md for details.