llama.cpp

lgrammel/ai-sdk-llama-cpp is a community provider that enables local LLM inference using llama.cpp directly within Node.js via native C++ bindings.

This provider loads llama.cpp directly into Node.js memory, eliminating the need for an external server while providing native performance and GPU acceleration.

Features

• Native Performance: Direct C++ bindings using node-addon-api (N-API)
• GPU Acceleration: Automatic Metal support on macOS
• Streaming & Non-streaming: Full support for both generateText and streamText
• Structured Output: Generate JSON objects with schema validation using Output
• Embeddings: Generate embeddings with embed and embedMany
• Chat Templates: Automatic or configurable chat template formatting (llama3, chatml, gemma, etc.)
• GGUF Support: Load any GGUF-format model

Prerequisites

Before installing, ensure you have the following:

• macOS (Apple Silicon or Intel)
• Node.js >= 22.0.0
• CMake >= 3.15
• Xcode Command Line Tools

# Install Xcode Command Line Tools (includes Clang)
xcode-select --install

# Install CMake via Homebrew
brew install cmake

Setup

The llama.cpp provider is available in the ai-sdk-llama-cpp module. You can install it with:

pnpm add ai-sdk-llama-cpp

npm install ai-sdk-llama-cpp

yarn add ai-sdk-llama-cpp

bun add ai-sdk-llama-cpp

The installation will automatically compile llama.cpp as a static library with Metal support and build the native Node.js addon.

Provider Instance

You can import llamaCpp from ai-sdk-llama-cpp and create a model instance:

import { llamaCpp } from 'ai-sdk-llama-cpp';

const model = llamaCpp({
  modelPath: './models/Llama-3.2-1B-Instruct-Q4_K_M.gguf',
});

Configuration Options

You can customize the model instance with the following options:

• modelPath string (required)

  Path to the GGUF model file.

• contextSize number

  Maximum context size. Default: 2048.

• gpuLayers number

  Number of layers to offload to GPU. Default: 99 (all layers). Set to 0 to disable GPU.

• threads number

  Number of CPU threads. Default: 4.

• debug boolean

  Enable verbose debug output from llama.cpp. Default: false.

• chatTemplate string

  Chat template to use for formatting messages. Default: "auto" (uses the template embedded in the GGUF model file). Available templates include: llama3, chatml, gemma, mistral-v1, mistral-v3, phi3, phi4, deepseek, and more.

const model = llamaCpp({
  modelPath: './models/your-model.gguf',
  contextSize: 4096,
  gpuLayers: 99,
  threads: 8,
  chatTemplate: 'llama3',
});

Language Models

Text Generation

You can use llama.cpp models to generate text with the generateText function:

import { generateText } from 'ai';
import { llamaCpp } from 'ai-sdk-llama-cpp';

const model = llamaCpp({
  modelPath: './models/Llama-3.2-1B-Instruct-Q4_K_M.gguf',
});

try {
  const { text } = await generateText({
    model,
    prompt: 'Explain quantum computing in simple terms.',
  });

  console.log(text);
} finally {
  await model.dispose();
}

Streaming

The provider fully supports streaming with streamText:

import { streamText } from 'ai';
import { llamaCpp } from 'ai-sdk-llama-cpp';

const model = llamaCpp({
  modelPath: './models/Llama-3.2-1B-Instruct-Q4_K_M.gguf',
});

try {
  const result = streamText({
    model,
    prompt: 'Write a haiku about programming.',
  });

  for await (const chunk of result.textStream) {
    process.stdout.write(chunk);
  }
} finally {
  await model.dispose();
}

Structured Output

Generate type-safe JSON objects that conform to a schema using Output:

import { generateText, Output } from 'ai';
import { z } from 'zod';
import { llamaCpp } from 'ai-sdk-llama-cpp';

const model = llamaCpp({
  modelPath: './models/your-model.gguf',
});

try {
  const { output: recipe } = await generateText({
    model,
    output: Output.object({
      schema: z.object({
        name: z.string(),
        ingredients: z.array(
          z.object({
            name: z.string(),
            amount: z.string(),
          }),
        ),
        steps: z.array(z.string()),
      }),
    }),
    prompt: 'Generate a recipe for chocolate chip cookies.',
  });

  console.log(recipe);
} finally {
  await model.dispose();
}

The structured output feature uses GBNF grammar constraints to ensure the model generates valid JSON that conforms to your schema.

Generation Parameters

Standard AI SDK generation parameters are supported:

const { text } = await generateText({
  model,
  prompt: 'Hello!',
  maxTokens: 256,
  temperature: 0.7,
  topP: 0.9,
  topK: 40,
  stopSequences: ['\n'],
});

Embedding Models

You can create embedding models using the llamaCpp.embedding() factory method:

import { embed, embedMany } from 'ai';
import { llamaCpp } from 'ai-sdk-llama-cpp';

const model = llamaCpp.embedding({
  modelPath: './models/nomic-embed-text-v1.5.Q4_K_M.gguf',
});

try {
  const { embedding } = await embed({
    model,
    value: 'Hello, world!',
  });

  const { embeddings } = await embedMany({
    model,
    values: ['Hello, world!', 'Goodbye, world!'],
  });
} finally {
  model.dispose();
}

Model Downloads

You'll need to download GGUF-format models separately. Popular sources:

• Hugging Face - Search for GGUF models
• TheBloke's Models - Popular quantized models

Example download:

# Create models directory
mkdir -p models

# Download a model (example: Llama 3.2 1B)
wget -P models/ https://huggingface.co/bartowski/Llama-3.2-1B-Instruct-GGUF/resolve/main/Llama-3.2-1B-Instruct-Q4_K_M.gguf

Resource Management

const model = llamaCpp({
  modelPath: './models/your-model.gguf',
});

try {
  // Use the model...
} finally {
  await model.dispose();
}

Limitations

• macOS only: Windows and Linux are not supported
• No tool/function calling: Tool calls are not supported
• No image inputs: Only text prompts are supported

Additional Resources

• GitHub Repository
• npm Package
• llama.cpp - The underlying inference engine