框架集成

在流行的 AI Agent 框架中使用 JarvisClaw。由于我们的 API 兼容 OpenAI，集成非常直接 — 大多数情况只需一行配置。

包

框架	包名	安装
LangChain	langchain-jarvisclaw	pip install langchain-jarvisclaw
CrewAI	crewai-jarvisclaw	pip install crewai-jarvisclaw
AutoGen (AG2)	autogen-jarvisclaw	pip install autogen-jarvisclaw

所有包都支持 API Key 和 x402 钱包支付两种模式。

LangChain

langchain-jarvisclaw 扩展了 ChatOpenAI，内置 x402 支持。

API Key

from langchain_jarvisclaw import ChatJarvisClaw

chat = ChatJarvisClaw(api_key="sk-...", model="gpt-5.4")
response = chat.invoke("解释量子计算")
print(response.content)

x402 钱包

from langchain_jarvisclaw import ChatJarvisClaw

# 按请求使用 USDC 支付 — 无需注册
chat = ChatJarvisClaw(wallet_private_key="0x...", model="gpt-5.4")
response = chat.invoke("解释量子计算")

链 & Agent

from langchain_core.prompts import ChatPromptTemplate
from langchain_jarvisclaw import ChatJarvisClaw

chat = ChatJarvisClaw(api_key="sk-...", model="anthropic/claude-sonnet-4.6")

prompt = ChatPromptTemplate.from_messages([
    ("system", "你是一个编程专家。"),
    ("human", "{question}"),
])

chain = prompt | chat
result = chain.invoke({"question": "如何反转链表？"})

流式输出

chat = ChatJarvisClaw(api_key="sk-...", model="gpt-5.4", streaming=True)

for chunk in chat.stream("给我讲个故事"):
    print(chunk.content, end="", flush=True)

发现（无需认证）

from langchain_jarvisclaw import ChatJarvisClaw

# 列出所有模型及 USD 定价
models = ChatJarvisClaw.list_models()
for m in models[:5]:
    print(f"{m['model']}: ${m['input_per_m_token_usd']}/M tokens")

# 查找免费模型
free = ChatJarvisClaw.free_models()

# 平台健康状态
health = ChatJarvisClaw.health()

CrewAI

crewai-jarvisclaw 提供可直接用于 CrewAI agent 的 LLM 类。

单 Agent

from crewai import Agent, Task, Crew
from crewai_jarvisclaw import JarvisClawLLM

llm = JarvisClawLLM(api_key="sk-...", model="gpt-5.4")

researcher = Agent(
    role="研究员",
    goal="查找关于 AI 趋势的信息",
    llm=llm,
)

task = Task(description="研究 2026 年排名前 5 的 AI 框架", agent=researcher)
crew = Crew(agents=[researcher], tasks=[task])
result = crew.kickoff()

x402 钱包

from crewai_jarvisclaw import JarvisClawLLM

# 按请求支付，无需 API key
llm = JarvisClawLLM(wallet_private_key="0x...", model="anthropic/claude-sonnet-4.6")

多 Agent

from crewai import Agent, Task, Crew
from crewai_jarvisclaw import JarvisClawLLM

# 不同 agent 可以使用不同模型
researcher_llm = JarvisClawLLM(api_key="sk-...", model="gpt-5.4")
writer_llm = JarvisClawLLM(api_key="sk-...", model="anthropic/claude-sonnet-4.6")

researcher = Agent(role="研究员", goal="查找信息", llm=researcher_llm)
writer = Agent(role="写作者", goal="撰写内容", llm=writer_llm)

research_task = Task(description="研究量子计算", agent=researcher)
write_task = Task(description="写一篇摘要", agent=writer)

crew = Crew(agents=[researcher, writer], tasks=[research_task, write_task])
result = crew.kickoff()

无需专用包

# CrewAI 原生支持 OpenAI 兼容端点：
from crewai import LLM
llm = LLM(
    model="openai/gpt-5.4",
    base_url="https://api.jarvisclaw.ai/v1",
    api_key="sk-...",
)

AutoGen (AG2)

autogen-jarvisclaw 为 AutoGen 的 config_list 格式提供配置帮助。

单模型

from autogen import ConversableAgent
from autogen_jarvisclaw import jarvisclaw_config

assistant = ConversableAgent(
    name="assistant",
    system_message="你是一个有帮助的 AI 助手。",
    llm_config={"config_list": [jarvisclaw_config(model="gpt-5.4", api_key="sk-...")]},
)

user = ConversableAgent(name="user", human_input_mode="NEVER", llm_config=False)
user.initiate_chat(assistant, message="法国的首都是哪里？")

多模型降级

from autogen import ConversableAgent
from autogen_jarvisclaw import jarvisclaw_config_list

# AutoGen 按顺序尝试模型 — 自动降级
configs = jarvisclaw_config_list(
    models=["gpt-5.4", "anthropic/claude-sonnet-4.6", "deepseek/deepseek-chat"],
    api_key="sk-...",
)

assistant = ConversableAgent(
    name="assistant",
    llm_config={"config_list": configs},
)

多 Agent

from autogen import ConversableAgent, GroupChat, GroupChatManager
from autogen_jarvisclaw import jarvisclaw_config

config = {"config_list": [jarvisclaw_config(model="gpt-5.4", api_key="sk-...")]}

researcher = ConversableAgent(name="researcher", system_message="研究主题。", llm_config=config)
writer = ConversableAgent(name="writer", system_message="撰写摘要。", llm_config=config)
critic = ConversableAgent(name="critic", system_message="审查准确性。", llm_config=config)

group_chat = GroupChat(agents=[researcher, writer, critic], messages=[], max_round=6)
manager = GroupChatManager(groupchat=group_chat, llm_config=config)
researcher.initiate_chat(manager, message="研究 2026 年量子计算")

无需专用包

# AutoGen 兼容 OpenAI — 直接配置也可：
config_list = [{
    "model": "gpt-5.4",
    "api_key": "sk-...",
    "base_url": "https://api.jarvisclaw.ai/v1",
}]

Eliza (ai16z)

Eliza agent 可以使用 JarvisClaw 作为模型供应商。在角色配置中添加：

{
  "modelProvider": "openai",
  "settings": {
    "model": "gpt-5.4",
    "apiKey": "sk-...",
    "baseURL": "https://api.jarvisclaw.ai/v1"
  }
}

任何 OpenAI 兼容框架

由于 JarvisClaw 完全兼容 OpenAI，任何允许设置 base_url 的框架都可以直接使用：

# 通用模式 — 适用于任何 OpenAI 兼容客户端
base_url = "https://api.jarvisclaw.ai/v1"
api_key = "sk-..."

无需特殊适配器。我们的包只是添加了 x402 钱包支付支持和便捷方法。