A lightweight Agent framework with native-level DeepSeek adaptation — Precise tool calling in thinking mode · Reliable structured output · Maximum cache hit rate.
Try it online — no local setup needed:
Or scaffold a project locally:
pnpx create deepseek-kit my-agentLangChain.js and AI SDK are excellent general-purpose frameworks, but DeepSeek's API has unique mechanisms that they cannot properly handle. deepseek-kit is built from the ground up to solve these problems.
DeepSeek's thinking mode outputs a chain of thought (reasoning_content) before the final answer. When the model makes tool calls during thinking, all subsequent requests must include the full reasoning_content — otherwise the API returns a 400 error.
General-purpose frameworks cannot distinguish between the different handling requirements for reasoning_content in "with tool call" vs "without tool call" scenarios, causing multi-turn tool calling to fail frequently.
deepseek-kit automatically tracks and re-sends reasoning_content in the agent loop, applies differentiated strategies based on whether tool calls occurred, and enables thinking mode by default — zero configuration needed.
DeepSeek API enables context hard disk caching by default. When subsequent requests have a prefix that exactly matches a previous request, the repeated portion is served from cache, significantly reducing latency and cost.
General-purpose frameworks often inject dynamic metadata (timestamps, request IDs) or arrange messages in non-deterministic order, breaking prefix consistency and causing cache hit rates to plummet.
deepseek-kit sends zero-redundancy request bodies with deterministic message construction, ensuring the same input always produces the same request prefix. Cache hit rates are fully observable via prompt_cache_hit_tokens and prompt_cache_miss_tokens.
Structured output is essential for agent applications, but under DeepSeek's thinking mode, general-purpose frameworks' structured output solutions often conflict with reasoning_content management, resulting in unreliable output formats.
deepseek-kit provides Zod Schema-driven structured output with smart retry and formatted error feedback, fully compatible with thinking mode — the chain-of-thought context is never lost during formatting steps.
- 🧠 Thinking Mode Adaptation — Automatic
reasoning_contentmanagement, zero-config tool calling chains - 💾 Cache Hit Rate Optimization — Zero-redundancy request body + deterministic message construction
- 📋 Structured Output — Zod Schema-driven, smart retry, thinking mode compatible
- 🤖 Agent System — Build intelligent agents with tool calling and multi-step execution
- 🌿 Subagents — Encapsulate agents as tools for delegation, with isolated context and parallel execution
- 💬 Streaming — Streaming events for text, chain-of-thought, and tool calls
- 🔧 Tool Calling — Built-in tool definition, parameter validation, timeout, and retry
- ✍️ FIM Completion — Fill-in-the-Middle code completion support
- 🪝 Hook System — Insert custom logic before and after generation steps
- 🔄 Auto Retry — Smart retry strategy with exponential backoff and jitter
- 🌲 Tree-shakable — Pure ESM,
sideEffects: false - 🔒 Type Safe — Complete TypeScript type definitions
pnpm add deepseek-kitimport { createAgent, createModel, tool } from 'deepseek-kit'
import { z } from 'zod'
const model = createModel({ model: 'deepseek-v4-flash' })
const weatherTool = tool({
name: 'get_weather',
description: 'Get weather information for a city',
schema: z.object({
city: z.string().describe('City name'),
}),
execute: async ({ city }) => `${city}: Sunny, 25°C`,
})
const agent = createAgent({ model, tools: [weatherTool] })
const result = await agent.generate({
prompt: 'How\'s the weather in Chongqing today?',
})
console.log(result.text)Requirements: Node.js >= 18.0.0, DeepSeek API key
📖 For the full guide, visit the documentation.
MIT License © Flippedround