智能体应用API调用

AgentApp 成功部署并监听 127.0.0.1:8090 后,可以通过流式接口、OpenAI 兼容接口和 AG-UI 接口完成推理调用。

环境要求

  • 已按照官方流程启动 AgentApp(示例名称为 Friday)。

  • 在运行环境中配置好模型密钥,例如 DASHSCOPE_API_KEY

  • 客户端推荐使用支持异步与 SSE 的 HTTP 库(如 aiohttp, httpx)。

/process:SSE 流式接口

请求体格式

{
  "input": [
    {
      "role": "user",
      "content": [
        { "type": "text", "text": "What is the capital of France?" }
      ]
    }
  ],
  "session_id": "可选,同一会话复用",
  "user_id": "可选,便于区分多用户"
}
  • input 遵循 Agentscope 消息格式,可包含多条消息及富媒体内容。

  • session_id 用于让 StateService / SessionHistoryService 等服务记录上下文,支持多轮记忆以及智能体状态持久化。

  • user_id 默认可缺省,如需统计不同账号可自行传入。

解析流式响应

服务端以 Server-Sent Events 协议返回增量结果,每行形如 data: {...},最终以 data: [DONE] 结束。下方代码节选自 test_process_endpoint_stream_async,演示如何解析并抽取文本:

async with session.post(url, json=payload) as resp:
    assert resp.headers["Content-Type"].startswith("text/event-stream")
    async for chunk, _ in resp.content.iter_chunks():
        if not chunk:
            continue
        line = chunk.decode("utf-8").strip()
        if not line.startswith("data:"):
            continue
        data_str = line[len("data:") :].strip()
        if data_str == "[DONE]":
            break
        event = json.loads(data_str)
        text = event["output"][0]["content"][0]["text"]
        # 在此累积或实时显示 text

多轮对话示例

下面的例子展示了如何在多次请求中复用 session_id,实现“记住用户姓名”等效果:

  1. 第一次调用:"My name is Alice."

  2. 第二次调用:"What is my name?"

SSE 输出中会包含 “Alice”,表明状态与会话历史已经生效。

session_id = "123456"

url = f"http://localhost:{PORT}/process"

async with aiohttp.ClientSession() as session:
    payload1 = {
        "input": [
            {
                "role": "user",
                "content": [{"type": "text", "text": "My name is Alice."}],
            },
        ],
        "session_id": session_id,
    }
    async with session.post(url, json=payload1) as resp:
        assert resp.status == 200
        assert resp.headers.get("Content-Type", "").startswith(
            "text/event-stream",
        )
        async for chunk, _ in resp.content.iter_chunks():
            if not chunk:
                continue
            line = chunk.decode("utf-8").strip()
            if (
                line.startswith("data:")
                and line[len("data:") :].strip() == "[DONE]"
            ):
                break

payload2 = {
    "input": [
        {
            "role": "user",
            "content": [{"type": "text", "text": "What is my name?"}],
        },
    ],
    "session_id": session_id,
}

async with aiohttp.ClientSession() as session:
    async with session.post(url, json=payload2) as resp:
        assert resp.status == 200
        assert resp.headers.get("Content-Type", "").startswith(
            "text/event-stream",
        )

        found_name = False

        async for chunk, _ in resp.content.iter_chunks():
            if not chunk:
                continue
            line = chunk.decode("utf-8").strip()
            if line.startswith("data:"):
                data_str = line[len("data:") :].strip()
                if data_str == "[DONE]":
                    break
                try:
                    event = json.loads(data_str)
                except json.JSONDecodeError:
                    continue

                if "output" in event:
                    try:
                        text_content = event["output"][0]["content"][0][
                            "text"
                        ].lower()
                        if "alice" in text_content:
                            found_name = True
                    except Exception:
                        pass

        assert found_name, "Did not find 'Alice' in the second turn output"

/compatible-mode/v1/responses:OpenAI 兼容接口

若现有系统已经接入 OpenAI 官方 SDK,可直接指向兼容端点,几乎零改造即可使用:

from openai import OpenAI

client = OpenAI(base_url="http://127.0.0.1:8090/compatible-mode/v1")
resp = client.responses.create(
    model="any_name",
    input="Who are you?",
)
print(resp.response["output"][0]["content"][0]["text"])

该接口完全复用了 Responses API 的返回格式,测试用例 test_openai_compatible_mode 也验证了模型会自报家门为 “Friday”。

/ag-ui:AG-UI Protocol兼容接口

AgentScope Runtime 通过适配器模式支持了AG-UI Protocol,基于 @app.query 装饰器构建的Agent API,会自动适配为AG-UI Protocol的请求体格式。

AG-UI 请求体格式

HTTP API 中,POST /ag-ui 的 body 即 RunAgentInput。

{
  "threadId": "thread_123",
  "runId": "run_456",
  "parentRunId": "run_000",
  "state": {},
  "messages": [
    {
      "id": "msg_1",
      "role": "user",
      "content": "你好,请介绍一下你自己"
    }
  ],
  "tools": [],
  "context": [],
  "forwardedProps": {}
}

字段说明

  • threadId(必需):会话线程ID,用于标识一个对话会话

  • runId(必需):本次运行ID,每次请求应使用唯一的ID

  • messages(必需):消息列表,支持多种消息类型:

    • role: "user":用户消息

    • role: "assistant":助手消息(可包含文本或工具调用)

    • role: "system":系统消息

    • role: "tool":工具执行结果消息

  • tools(可选):可用工具列表

  • context(可选):上下文信息

  • state(可选):状态信息

  • forwardedProps(可选):转发属性

API响应

服务端返回 SSE(Server-Sent Events)格式的流式响应,每个事件以 data: 开头,包含 JSON 格式的事件数据。目前适配器支持的事件类型包括:

事件类型

事件类型

说明

示例字段

run_started

运行开始

thread_id, run_id

text_message_start

文本消息开始

message_id

text_message_content

文本消息增量内容

message_id, delta

text_message_end

文本消息结束

message_id

tool_call_start

工具调用开始

tool_call_id, tool_call_name

tool_call_args

工具调用参数

tool_call_id, delta

tool_call_end

工具调用结束

tool_call_id

tool_call_result

工具执行结果

tool_call_id, content

run_finished

运行完成

thread_id, run_id

run_error

运行错误

run_id, message, code

调用示例

import aiohttp
import json

async def call_ag_ui_endpoint():
    url = "http://localhost:8080/ag-ui"
    payload = {
        "threadId": "thread_1234",
        "runId": "run_4567",
        "messages": [
            {
                "id": "msg_1",
                "role": "user",
                "content": "北京今天的天气如何"
            }
        ]
    }

    async with aiohttp.ClientSession() as session:
        async with session.post(url, json=payload) as resp:
            assert resp.status == 200
            assert resp.headers["Content-Type"].startswith("text/event-stream")

            async for chunk, _ in resp.content.iter_chunks():
                if not chunk:
                    continue

                line = chunk.decode("utf-8").strip()
                if not line.startswith("data:"):
                    continue
                print(line + "\n")



返回示例:

data: {"type": "RUN_STARTED", "thread_id": "thread_1234", "run_id": "run_4567"}

data: {"type": "TOOL_CALL_START", "tool_call_id": "call_c51915f8d0ab4c6aac85e1", "tool_call_name": "get_weather", "message_id": "msg_57562f7d-e96d-4de9-8899-1334fc101e23_0"}

data: {"type": "TOOL_CALL_ARGS", "tool_call_id": "call_c51915f8d0ab4c6aac85e1", "delta": "{\"location\": \"北京\"}"}

data: {"type": "TOOL_CALL_END", "tool_call_id": "call_c51915f8d0ab4c6aac85e1"}

data: {"type": "TOOL_CALL_RESULT", "message_id": "msg_0ca9a23b-0674-496b-91c8-5bd699945e70_0", "tool_call_id": "call_c51915f8d0ab4c6aac85e1", "content": "[{\"type\": \"text\", \"text\": \"The weather in 北京 is sunny with a temperature of 25°C.\"}]", "role": "tool"}

data: {"type": "TEXT_MESSAGE_START", "message_id": "msg_8debb51f-3226-4f1a-a573-5f80db132f80_0", "role": "assistant"}

data: {"type": "TEXT_MESSAGE_CONTENT", "message_id": "msg_8debb51f-3226-4f1a-a573-5f80db132f80_0", "delta": "北京"}

data: {"type": "TEXT_MESSAGE_CONTENT", "message_id": "msg_8debb51f-3226-4f1a-a573-5f80db132f80_0", "delta": "今天的天气是"}

data: {"type": "TEXT_MESSAGE_CONTENT", "message_id": "msg_8debb51f-3226-4f1a-a573-5f80db132f80_0", "delta": "晴朗,"}

data: {"type": "TEXT_MESSAGE_CONTENT", "message_id": "msg_8debb51f-3226-4f1a-a573-5f80db132f80_0", "delta": "气温为25"}

data: {"type": "TEXT_MESSAGE_CONTENT", "message_id": "msg_8debb51f-3226-4f1a-a573-5f80db132f80_0", "delta": "°C。"}

data: {"type": "TEXT_MESSAGE_END", "message_id": "msg_8debb51f-3226-4f1a-a573-5f80db132f80_0"}

data: {"type": "RUN_FINISHED", "thread_id": "thread_1234", "run_id": "run_4567"}

前端集成

基于 CopilotKit 可以快速基于已有的 AG-UI Protocol API 构建 Agent 应用,详细可以参考 CopilotKit文档

常见排查

  • 无法连接:确认服务仍在运行、端口未被占用,并检查客户端是否访问了正确的主机(容器或远程部署时尤为重要)。

  • 解析失败:SSE 帧需要逐行处理;如出现 keep-alive 空行或半截 JSON,请加上容错逻辑。

  • 不记得上下文:检查是否在所有请求中传入相同的 session_id,以及 state/session 服务是否在 init_func 中正确 start()