kimi-vercel-ai-sdk-provider weekly downloads statistics

kimi-vercel-ai-sdk-provider

<img width="1500" height="762" alt="git" src="https://github.com/user-attachments/assets/5b8d9be4-5eb7-46f3-9096-56bff8dcda67" />

This is a native implementation with full support for Kimi-specific features, not a generic OpenAI-compatible wrapper.

Table of Contents

• Features
• Install
• Quick Start
• Kimi Code
  • Available Models
  • Extended Thinking
  • Streaming with Thinking Blocks
• Built-in Tools
  • Web Search
  • Code Interpreter
  • Provider Tool Helpers
• Native File & PDF Support
  • File Client
  • Attachment Processing
  • Supported File Types
• Tool Choice Polyfill
• Context Caching
• Token Tracking
• Reasoning/Thinking Models
• Video Input
• Model Capabilities
• Advanced Features
  • Auto-Detect Tools
  • Ensemble Generation
  • Code Validation
  • Multi-Agent Collaboration
  • Project Scaffolding
  • Temperature Locking
  • File Content Caching
  • Schema Sanitization
  • Reasoning Preservation
• Provider Options
• Available Models
• Regional Endpoints
• Environment Variables
• Why Native vs OpenAI-Compatible?
• API Reference
• License

Features

Kimi Chat (Standard API)

• Built-in web search ($web_search) for grounded responses
• Built-in code interpreter ($code) for programmatic reasoning
• Deep thinking/reasoning model streaming
• Video input support for K2.5 models
• 256k context window support
• Token tracking (cache hits, reasoning, web search, code interpreter)
• Regional endpoints (global and China)
• Provider tool helpers (kimi.tools.* and kimiTools.*)
• Native File & PDF Support - Automatic file upload and content extraction
• Tool Choice Polyfill - Simulates required and tool choices via system messages
• Context Caching - Reduce costs by up to 90% for repeated long prompts
• Temperature Locking - Automatic temperature enforcement for thinking models
• File Content Caching - LRU cache to avoid re-uploading identical files
• Schema Sanitization - Automatic cleanup of unsupported JSON Schema keywords

Kimi Code (Premium Coding API)

• High-speed output (up to 100 tokens/s)
• Extended thinking/reasoning support with configurable effort levels
• 262k context window
• Streaming with thinking blocks

Install

npm install kimi-vercel-ai-sdk-provider

Quick Start

Kimi Chat (Standard)

import { createKimi } from 'kimi-vercel-ai-sdk-provider';
import { generateText, streamText } from 'ai';

const kimi = createKimi({
  // Uses MOONSHOT_API_KEY env var if not provided
  endpoint: 'global',
});

// Basic usage
const result = await generateText({
  model: kimi('kimi-k2.5'),
  prompt: 'Explain quantum computing in one sentence.',
});

// Streaming with reasoning model
const stream = await streamText({
  model: kimi('kimi-k2.5-thinking', { includeUsageInStream: true }),
  prompt: 'Solve this step by step: What is 17 * 23?',
});

Kimi Code (Premium)

import { kimiCode, createKimiCode } from 'kimi-vercel-ai-sdk-provider
';
import { generateText, streamText } from 'ai';

// Using default instance (uses KIMI_CODE_API_KEY or KIMI_API_KEY env var)
const result = await generateText({
  model: kimiCode(), // Uses 'kimi-for-coding' by default
  prompt: 'Write a TypeScript function to merge two sorted arrays',
});

// With extended thinking enabled
const stream = await streamText({
  model: kimiCode('kimi-k2-thinking', {
    extendedThinking: {
      enabled: true,
      effort: 'high' // 'low' | 'medium' | 'high'
    }
  }),
  prompt: 'Design a distributed cache system',
});

// Or with custom configuration
const customKimiCode = createKimiCode({
  apiKey: 'sk-kimi-xxx',
  baseURL: 'https://api.kimi.com/coding/v1', // default
});

Kimi Code

Kimi Code is a premium coding service optimized for development tasks with high-speed output and extended thinking support.

Available Models

Model	Description
kimi-for-coding	Primary coding model optimized for development tasks (default)
kimi-k2-thinking	Extended thinking model for complex reasoning

Extended Thinking

Enable extended thinking to see the model's reasoning process:

// Simple boolean
const model = kimiCode('kimi-for-coding', { extendedThinking: true });…