Skip to main content

A Python-native Agent framework with first-class Skill support and multi-LLM adapter

Project description

AgentKit

Python 原生 Agent 框架,内置一等公民 Skill 支持与多 LLM 适配器。

Python 3.11+ License: MIT

✨ 特性

  • 🤖 声明式 Agent — 零继承配置,支持 Handoff 转介 + as_tool 委派两种协作模式
  • 📚 一等公民 Skill — 三级渐进式加载(L1 元数据 → L2 指令 → L3 资源),按需加载省 token
  • 🔧 灵活工具系统@function_tool 装饰器自动推断 JSON Schema
  • 🧠 多 LLM 适配器 — 自研统一适配层,4 个适配器覆盖所有主流 LLM
  • 🛡️ 内置安全 — Guardrail 护栏 + 权限控制 + 三级沙箱
  • 🎭 编排 Agent — Sequential / Parallel / Loop 三种模式
  • 💾 记忆系统 — 可选集成 Mem0,支持自定义记忆提供者
  • 全异步设计与 Hooks — 底层全面采用 asyncio,流式响应(Streaming)、事件驱动(Event-Driven)、断点续跑机制(Checkpoint/Resume)。提供细粒度的 生命周期 Hooks,支持改写请求与结果。
  • 🏢 多租户数据隔离 — 框架级支持 user_id / session_id 贯穿,Memory 默认分桶,Session 结束自动释放资源。
  • 🔗 与大模型平台解耦 — 内置多模型支持,轻松对接 Ollama 等本地模型。

🚀 安装

# 基础安装
pip install ni.agentkit

# 如果需要 OpenAI / DeepSeek / 通义千问等
pip install "ni.agentkit[openai]"
pip install "ni.agentkit[anthropic]"
pip install "ni.agentkit[google]"
pip install "ni.agentkit[memory]"
pip install "ni.agentkit[docker]"
pip install "ni.agentkit[all]"

⚡ 30 秒快速开始

from agentkit import Agent, Runner, function_tool

@function_tool
def get_weather(city: str) -> str:
    """获取天气"""
    return f"{city}:晴,25°C"

agent = Agent(
    name="assistant",
    instructions="你是一个有帮助的中文助手。",
    model="ollama/qwen3.5:cloud",
    tools=[get_weather],
)

result = Runner.run_sync(agent, input="北京今天天气如何?")
print(result.final_output)

📚 文档

安装后查看文档:

# 命令行方式
agentkit-docs

# Python 方式
import agentkit
print(agentkit.get_docs_dir())     # 文档目录路径
print(agentkit.get_examples_dir()) # 示例目录路径
文档 说明
README 项目概述与特性
QuickStart 8 个渐进式入门示例
Architecture 六层架构设计说明
Reference 完整 API 参考手册

🧪 示例

安装包内含 16 个可运行示例(标准版 × 8 + Ollama 本地版 × 8):

# Ollama 本地版(无需 API Key)
python -c "import agentkit; print(agentkit.get_examples_dir())"
# 然后运行对应目录下的示例文件

# 或者直接:
python -m agentkit.examples.ollama.01_basic_chat

🔌 支持的 LLM

模型 适配器 用法
GPT-4o / o1 / o3 / o4 OpenAIAdapter model="gpt-4o"
Claude Opus/Sonnet/Haiku AnthropicAdapter model="claude-sonnet-4-20250514"
Gemini 2.5 / 3 GoogleAdapter model="gemini-2.5-pro"
通义千问/智谱/DeepSeek/Moonshot/百川/Azure OpenAICompatibleAdapter model="deepseek/deepseek-chat"
Ollama 本地模型 OllamaAdapter model="ollama/qwen3.5:cloud"

🔨 构建打包

./build.sh          # 构建 wheel + sdist
./build.sh clean    # 清理构建产物
./build.sh test     # 在隔离环境中安装并验证
./build.sh all      # 清理 + 构建 + 验证(推荐)

构建产物输出到 dist/ 目录:

dist/
├── ni_agentkit-0.3.2-py3-none-any.whl   # pip install 用这个
└── ni_agentkit-0.3.2.tar.gz             # 源码分发

📄 License

MIT

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

ni_agentkit-0.4.0.tar.gz (111.9 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

ni_agentkit-0.4.0-py3-none-any.whl (156.8 kB view details)

Uploaded Python 3

File details

Details for the file ni_agentkit-0.4.0.tar.gz.

File metadata

  • Download URL: ni_agentkit-0.4.0.tar.gz
  • Upload date:
  • Size: 111.9 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.13.2

File hashes

Hashes for ni_agentkit-0.4.0.tar.gz
Algorithm Hash digest
SHA256 49b3aec8dcff625d3b9675e9f23b3091f2517b65f0298231863ec4f1a2f3a74c
MD5 98a122cac5df2a6cc18b0c276067c990
BLAKE2b-256 906ca5f0b74f9cbee49894de6178c5dddfe394fe8713a2c92a641066739b5f7a

See more details on using hashes here.

File details

Details for the file ni_agentkit-0.4.0-py3-none-any.whl.

File metadata

  • Download URL: ni_agentkit-0.4.0-py3-none-any.whl
  • Upload date:
  • Size: 156.8 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.13.2

File hashes

Hashes for ni_agentkit-0.4.0-py3-none-any.whl
Algorithm Hash digest
SHA256 846411617e3380aee0a27ce0c52e75088290b3ba6862146d8dff48e8bb073b6f
MD5 e8ab433a9b52138d3204207a66cb76db
BLAKE2b-256 d6f4a0ac24d916c54606671b3e331c52c69019389a4069aa095697b2bfd041da

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page