Chat History

显式历史管理保证多轮可控。

源码

from agently import Agently

agent = Agently.create_agent()

Agently.set_settings(
    "OpenAICompatible",
    {
        "base_url": "http://127.0.0.1:11434/v1",
        "model": "qwen2.5:7b",
    },
)


## Chat History: basic multi-turn management
def chat_history_basic():
    # You can add messages to chat_history to keep multi-turn context.
    agent.set_chat_history(
        [
            {"role": "user", "content": "Hi, who are you?"},
            {"role": "assistant", "content": "I'm an Agently assistant."},
        ]
    )
    result = agent.input("What did I ask you before?").start()
    print(result)

    # You can append new turns, or reset the history.
    # Treat the last answer as a new user message to continue the thread.
    agent.add_chat_history({"role": "user", "content": result})
    follow_up = agent.input("Summarize my last message in one sentence.").start()
    print(follow_up)

    agent.reset_chat_history()


# chat_history_basic()

讲解

• 使用 set/add/reset 管理对话。
• 只保留必要轮次。

注释解读

• Idea 表示案例思路
• Flow 表示执行编排路径
• Expect 表示预期输出或行为

你学会了什么

• 显式管理历史对话以保证可控
• 理解 history 与 system prompt 的区别

练习任务

• 构造 3 轮对话并手动注入历史
• 只保留最近两轮观察变化