LLM Module

Framework abstraction to interact with LLM models, including prompt management and structured generation.

Quick Start

import { generateText } from '@output.ai/llm';

const response = await generateText({
  prompt: 'my_prompt@v1',
  variables: { topic: 'AI workflows' }
});

Features

• Unified API: Single import for prompt loading and LLM generation
• Multiple Generation Types: Text, objects, arrays, and enums
• Prompt Management: Load and render .prompt files with variable interpolation
• Multi-Provider Support: Anthropic, OpenAI, and Azure
• Type Safety: Full TypeScript support with Zod schemas

Generate Text

Generate unstructured text from an LLM:

import { generateText } from '@output.ai/llm';

const response = await generateText({
  prompt: 'explain_topic@v1',
  variables: { topic: 'machine learning' }
});

Generate Object

Generate a structured object matching a Zod schema:

import { generateObject } from '@output.ai/llm';
import { z } from '@output.ai/core';

const recipeSchema = z.object({
  title: z.string(),
  ingredients: z.array(z.string()),
  steps: z.array(z.string())
});

const recipe = await generateObject({
  prompt: 'recipe@v1',
  variables: { dish: 'lasagna' },
  schema: recipeSchema
});

Generate Array

Generate an array of structured items:

import { generateArray } from '@output.ai/llm';
import { z } from '@output.ai/core';

const taskSchema = z.object({
  title: z.string(),
  priority: z.number()
});

const tasks = await generateArray({
  prompt: 'task_list@v1',
  variables: { project: 'website' },
  schema: taskSchema
});

Generate Enum

Generate a value from a list of allowed options:

import { generateEnum } from '@output.ai/llm';

const category = await generateEnum({
  prompt: 'categorize@v1',
  variables: { text: 'Product announcement' },
  enum: ['marketing', 'engineering', 'sales', 'support']
});

Prompt Files

Prompt files use YAML frontmatter for configuration and support LiquidJS templating:

File: [email protected]

---
provider: anthropic
model: claude-sonnet-4-20250514
temperature: 0.7
---

<system>
You are a concise technical explainer.
</system>

<user>
Explain {{ topic }} in 3 bullet points.
</user>

Configuration Options

Prompt files support these configuration fields:

---
provider: anthropic | openai | azure
model: model-name
temperature: 0.0-1.0 (optional)
maxTokens: number (optional)
providerOptions: (optional)
  thinking:
    type: enabled
    budgetTokens: number
---