ApplicationContext JSON Serialization Error #755
Description
I encountered a JSON serialization error related to ApplicationContext when running the example code to create an example agent. The error occurs in the _save_trajectories method when attempting to serialize task trajectories, and appears to stem from the aworld.utils.serialized_util.to_serializable function not properly handling ApplicationContext objects.
Error Log
Failed to get trajectories: Object of type ApplicationContext is not JSON serializable.Traceback (most recent call last):
File "/Users/liyun/Downloads/AWorld-main/aworld/runners/event_runner.py", line 444, in _save_trajectories
"trajectory": json.dumps(self._task_response.trajectory, ensure_ascii=False),
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/Users/liyun/miniconda3/lib/python3.11/json/__init__.py", line 238, in dumps
**kw).encode(obj)
^^^^^^^^^^^
File "/Users/liyun/miniconda3/lib/python3.11/json/encoder.py", line 200, in encode
chunks = self.iterencode(o, _one_shot=True)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/Users/liyun/miniconda3/lib/python3.11/json/encoder.py", line 258, in iterencode
return _iterencode(o, 0)
^^^^^^^^^^^^^^^^^
File "/Users/liyun/miniconda3/lib/python3.11/json/encoder.py", line 180, in default
raise TypeError(f'Object of type {o.__class__.__name__} '
TypeError: Object of type ApplicationContext is not JSON serializable My code
from aworld.agents.llm_agent import Agent
from aworld.config.conf import AgentConfig, ModelConfig
from aworld.core.tool.func_to_tool import be_tool
from aworld.runner import Runners
from aworld.core.task import Task
from aworld.config.conf import AgentMemoryConfig
from aworld.core.memory import LongTermConfig
from aworld.core.context.amni.config import init_middlewares, AmniConfigFactory, AmniConfigLevel
from aworld.memory.main import MemoryFactory, InMemoryMemoryStore
from aworld.memory.models import AgentExperience
import asyncio
import os
# ---------------------- 步骤1：配置LLM（优先使用环境变量，安全且易部署） ----------------------
os.environ["LLM_PROVIDER"] = "openai"
os.environ["LLM_MODEL_NAME"] = "gpt-4"
os.environ["LLM_API_KEY"] = ""
os.environ["LLM_BASE_URL"] = "https://api.openai.com/v1"
# 构建共享ModelConfig（后续新增Agent可直接复用该配置）
model_config = ModelConfig(
llm_provider=os.getenv("LLM_PROVIDER", "openai"),
llm_model_name=os.getenv("LLM_MODEL_NAME"),
llm_base_url=os.getenv("LLM_BASE_URL"),
llm_api_key=os.getenv("LLM_API_KEY"),
)
init_middlewares()
# 构建Agent专属配置（含记忆配置）
agent_memory_config = AgentMemoryConfig(
enable_summary=True,
summary_rounds=5,
summary_context_length=8000,
history_rounds=20,
enable_long_term=True,
long_term_config=LongTermConfig.create_simple_config(
application_id="aworld-agent-example",
enable_user_profiles=True,
enable_agent_experiences=True
)
)
agent_config = AgentConfig(
llm_config=model_config,
memory_config=agent_memory_config,
)
# ---------------------- 步骤2：定义本地工具（Agent的核心能力扩展） ----------------------
# 工具1：基础问候工具（支持自定义用户名）
@be_tool(tool_name="custom_greeting_tool",tool_desc="A tool to generate a warm greeting message, accepts a user name parameter to personalize the greeting.")
def custom_greeting_tool(user_name: str = "User") -> str:
"""
生成个性化问候语
:param user_name: 用户名，可选参数，默认值为User
:return: 个性化问候字符串
"""
return f"Hello {user_name}! Nice to meet you, I'm your intelligent text processing assistant."
# 工具2：文本摘要工具（提炼文本核心信息）
@be_tool(tool_name="text_summarization_tool",tool_desc="A tool to summarize long text, extract the core information and key points, return a concise summary.")
def text_summarization_tool(text: str) -> str:
"""
对长文本进行摘要提取（依赖LLM的内置能力，此处封装为工具方便Agent调用）
:param text: 需要被摘要的长文本
:return: 简洁的文本摘要
"""
# 这里直接封装摘要逻辑，Agent会结合LLM优化输出结果
summary_prompt = f"Please summarize the following text concisely, within 100 words: \n\n{text}"
return summary_prompt
# 工具3：中英互译工具（支持中文转英文、英文转中文）
@be_tool(tool_name="cn_en_translate_tool",tool_desc="A tool for Chinese-English mutual translation, accepts text and target language parameters, target language options: 'zh' or 'en'.")
def cn_en_translate_tool(text: str, target_language: str = "en") -> str:
"""
中英互译工具
:param text: 需要翻译的文本
:param target_language: 目标语言，可选值：zh（中文）、en（英文），默认值为en
:return: 翻译提示词（Agent结合LLM输出最终翻译结果）
"""
if target_language not in ["zh", "en"]:
return "Error: Target language only supports 'zh' (Chinese) or 'en' (English)."
translate_prompt = f"Please translate the following text into {target_language}: \n\n{text}"
return translate_prompt
# ---------------------- 步骤3：设置System Prompt（定义Agent的角色和行为逻辑） ----------------------
system_prompt = """
You are an intelligent text processing assistant, named 'TextProAgent'.
Your core capabilities are provided by three tools, and you must use the corresponding tools to complete user tasks:
1. Use custom_greeting_tool when the user needs a greeting or welcome message.
2. Use text_summarization_tool when the user needs to summarize long text, extract key points.
3. Use cn_en_translate_tool when the user needs Chinese-English mutual translation.
Rules to follow:
- Be polite and clear in your response.
- When using tools, follow the tool's parameter requirements.
- If the user's request is not related to your capabilities, politely remind the user of your core functions.
"""
# ---------------------- 步骤4：实例化Agent（整合所有配置和工具） ----------------------
text_pro_agent = Agent(
name="TextProAgent", # Agent的唯一名称
conf=agent_config, # 之前配置的LLM和Agent参数
system_prompt=system_prompt, # 定义Agent角色的系统提示词
tool_names=[ # 注册当前Agent需要使用的本地工具（与上面定义的工具名一致）
"custom_greeting_tool",
"text_summarization_tool",
"cn_en_translate_tool"
]
)
# ---------------------- 步骤5：验证Agent（可选，用于测试Agent是否创建成功） ----------------------
print("智能文本处理Agent创建成功！")
print(f"Agent名称：{text_pro_agent.name}")
print(f"已加载工具数量：{len(text_pro_agent.tool_names)}")
# ---------------------- 步骤6：通过 Runners 调用 Agent 执行具体任务 ----------------------
RUN_AGENT_TASKS = True # 不依赖模型连接时，设为 False
def run_task(task_input: str, user_id: str = "user_001", session_id: str = "session_001") -> str:
"""统一用运行时入口执行，保证上下文/记忆/轨迹等能力生效。"""
context_config = AmniConfigFactory.create(level=AmniConfigLevel.PILOT)
task = Task(
input=task_input,
agent=text_pro_agent,
user_id=user_id,
session_id=session_id,
context_config=context_config
)
result = Runners.sync_run_task(task=task)
return result.get(task.id).answer
if RUN_AGENT_TASKS:
print("\n" + "="*50)
print("开始调用 Agent 执行任务...")
print("="*50)
# 任务1：测试 个性化问候工具（传入用户名）
print("\n【任务1：个性化问候】")
greeting_task = "请给我一个问候，我的名字叫小李"
greeting_result = run_task(greeting_task)
print("Agent 回复：", greeting_result)