larry-sdk

TypeScript SDK for interacting with the Larry-as-a-Service API. It handles generator discovery, AI-assisted schema extraction via a streaming chat conversation, and code generation.

Usage

The SDK requires an API key. Provide it via the constructor or the LARRY_API_KEY environment variable.

import { LarrySDK } from '@codestrap/larry-sdk';

const sdk = new LarrySDK({
  apiKey: 'your-api-key', // or set LARRY_API_KEY env var
  apiUrl: 'https://...', // optional, defaults to https://larry-as-a-service.vercel.app
  rootDir: '/path/to/workspace', // optional, defaults to Nx workspaceRoot. It will be used to discover `package.json` of your project and adding missing dependencies required by generators
});

API

getGenerators(): Promise<Generator[]>

Fetches the list of available generators from the registry.

const generators = await sdk.getGenerators();
// [{ id, description, type, package, schema, exampleSchema }, ...]

generateCode(generatorId, schemaData, outputPath, onEmit)

Runs a generator against the in-memory file tree.

await sdk.generateCode(
  'ReactComponentGenerator',
  { name: 'Button', ui: 'shadcn' },
  'src/components/Button',
  (type: string, payload: Record<string, unknown>) =>
    console.log([type, payload]),
);

| Parameter | Type | Description | | ------------- | ----------------------------------------------------------- | --------------------------------------- | | generatorId | keyof typeof generators | ID of the generator to run | | schemaData | string \| Record<string, unknown> | Schema input for the generator | | outputPath | string | Output path written into the file tree | | onEmit | (type: string, payload?: Record<string, unknown>) => void | Callback for generator lifecycle events |

spawnExtractSchemaConversation(generator, onChunk?)

Opens a multi-turn chat conversation that helps a user describe what they want to generate. When the model determines it has enough information, it automatically extracts a structured schema and closes the conversation.

Returns an object with a single sendMessage(msg) method.

const convo = sdk.spawnExtractSchemaConversation(generator, (delta) => {
  process.stdout.write(delta); // stream response tokens
});

const first = await convo.sendMessage('I want to generate a Button component');
// { hasFinished: false, response: '...' }

const second = await convo.sendMessage(
  'It should accept a label and onClick prop',
);
// { hasFinished: false, response: '...' }
//   — or —
// { hasFinished: true, response: '...', generatedSchema: { ... } }

Once hasFinished is true, the generatedSchema field contains the extracted schema ready to pass to generateCode. Calling sendMessage again after the conversation finishes throws an error.

getFsTree(): FsTree

Returns the current in-memory Nx file tree. Use this to inspect or flush generated files.

const tree = sdk.getFsTree();

createFsTree(root, isDryRun, ...)

Replaces the current file tree with a new FsTree instance. Accepts the same arguments as the FsTree constructor.

sdk.createFsTree('/some/other/root', false);

Environment Variables

| Variable | Description | | --------------- | --------------------------------------------------- | | LARRY_API_KEY | API key (required if not passed to the constructor) | | LARRY_API_URL | API base URL (optional, overrides the default) |