AgentQL 集成

使用 LangChain Python 与 AgentQL 工具集成。

AgentQL 工具提供网页交互和从任何网页提取结构化数据的功能，使用 AgentQL 查询 或自然语言提示词。AgentQL 可用于多种语言和网页，且不会随时间推移或变化而失效。

概述

AgentQL 提供以下三种工具：

• ExtractWebDataTool 根据 URL 从网页中提取结构化 JSON 数据，使用 AgentQL 查询 或对数据的自然语言描述。

以下两个工具也打包为 AgentQLBrowserToolkit，必须与 Playwright 浏览器或通过 Chrome DevTools 协议 (CDP) 的远程浏览器实例一起使用：

• ExtractWebDataBrowserTool 从浏览器中的活动网页中提取结构化 JSON 数据，使用 AgentQL 查询 或自然语言描述。

• GetWebElementBrowserTool 使用自然语言描述在浏览器的活动网页上查找 Web 元素，并返回其 CSS 选择器以便进一步交互。

集成详情

类	包	可序列化	JS 支持	版本
AgentQL	langchain-agentql	❌	❌	1.0.0

工具特性

工具	Web 数据提取	Web 元素提取	与本地浏览器配合使用
ExtractWebDataTool	✅	❌	❌
ExtractWebDataBrowserTool	✅	❌	✅
GetWebElementBrowserTool	❌	✅	✅

设置

pip install --quiet -U langchain-agentql

要运行此笔记本，请安装 Playwright 浏览器并配置 Jupyter Notebook 的 asyncio 循环。

!playwright install

# This import is required only for jupyter notebooks, since they have their own eventloop
import nest_asyncio

nest_asyncio.apply()

凭据

要使用 AgentQL 工具，您需要从 AgentQL 开发者门户 获取自己的 API 密钥，并设置 AgentQL 环境变量。

import os

os.environ["AGENTQL_API_KEY"] = "YOUR_AGENTQL_API_KEY"

实例化

ExtractWebDataTool

您可以使用以下参数实例化 ExtractWebDataTool：

• api_key: 您的 AgentQL API 密钥，来自 dev.agentql.com。 可选。
• timeout: 请求超时前等待的秒数。如果数据提取超时，请增加此值。 默认为 900。
• is_stealth_mode_enabled: 是否启用实验性反爬虫规避策略。此功能可能无法在所有时间对所有网站生效。启用此模式后，数据提取可能需要更长时间才能完成。 默认为 False。
• wait_for: 提取数据前等待页面加载的秒数。 默认为 0。
• is_scroll_to_bottom_enabled: 是否在提取数据前滚动到页面底部。 默认为 False。
• mode: "standard" 使用深度数据分析，而 "fast" 以牺牲部分分析深度换取速度，适用于大多数用例。在此指南中了解更多关于模式的信息。 默认为 "fast"。
• is_screenshot_enabled: 是否在提取数据前截图。作为 Base64 字符串返回在 ‘metadata’ 中。 默认为 False。

ExtractWebDataTool 使用 AgentQL 的 REST API 实现，您可以在 API 参考文档 中查看有关参数的更多详细信息。

from langchain_agentql.tools import ExtractWebDataTool

extract_web_data_tool = ExtractWebDataTool()

ExtractWebDataBrowserTool

要实例化 ExtractWebDataBrowserTool，您需要将工具连接到浏览器实例。

您可以设置以下参数：

• timeout: 请求超时前等待的秒数。如果数据提取超时，请增加此值。 默认为 900。
• wait_for_network_idle: 是否在执行前等待网络达到完全空闲状态。 默认为 True。
• include_hidden: 是否考虑页面上视觉上隐藏的元素。 默认为 True。
• mode: "standard" 使用深度数据分析，而 "fast" 以牺牲部分分析深度换取速度，适用于大多数用例。在此指南中了解更多关于模式的信息。 默认为 "fast"。

ExtractWebDataBrowserTool 使用 AgentQL 的 SDK 实现。您可以在 AgentQL 的 API 参考 中找到有关参数和函数的更多详细信息。

from langchain_agentql.tools import ExtractWebDataBrowserTool
from langchain_agentql.utils import create_async_playwright_browser

async_browser = await create_async_playwright_browser()

extract_web_data_browser_tool = ExtractWebDataBrowserTool(async_browser=async_browser)

GetWebElementBrowserTool

要实例化 GetWebElementBrowserTool，您需要将工具连接到浏览器实例。

您可以设置以下参数：

• timeout: 请求超时前等待的秒数。如果数据提取超时，请增加此值。 默认为 900。
• wait_for_network_idle: 是否在执行前等待网络达到完全空闲状态。 默认为 True。
• include_hidden: 是否考虑页面上视觉上隐藏的元素。 默认为 False。
• mode: "standard" 使用深度数据分析，而 "fast" 以牺牲部分分析深度换取速度，适用于大多数用例。在此指南中了解更多关于模式的信息。 默认为 "fast"。

GetWebElementBrowserTool 使用 AgentQL 的 SDK 实现。您可以在 AgentQL 的 API 参考 中找到有关参数和函数的更多详细信息。

from langchain_agentql.tools import GetWebElementBrowserTool

extract_web_element_tool = GetWebElementBrowserTool(async_browser=async_browser)

调用

ExtractWebDataTool

此工具底层使用 AgentQL 的 REST API，将公开可用的网页 URL 发送到 AgentQL 的端点。这不适用于私有页面或已登录会话。对于这些用例，请使用 ExtractWebDataBrowserTool。

• url: 您要从中提取数据的网页 URL。
• query: 要执行的 AgentQL 查询。如果您想提取精确的结构化数据，请使用 AgentQL 查询。了解如何在文档中 编写 AgentQL 查询，或在 AgentQL 游乐场 中测试一个。
• prompt: 要从页面提取的数据的自然语言描述。AgentQL 将根据您的提示词推断数据结构。如果您想提取由自由格式语言定义的数据而不定义特定结构，请使用 prompt。

注意： 您必须定义 query 或 prompt 才能使用 AgentQL。

# You can invoke the tool with either a query or a prompt

# extract_web_data_tool.invoke(
#     {
#         "url": "https://www.agentql.com/blog",
#         "prompt": "the blog posts with title, url, date of post and author",
#     }
# )

extract_web_data_tool.invoke(
    {
        "url": "https://www.agentql.com/blog",
        "query": "{ posts[] { title url date author } }",
    },
)

{'data': {'posts': [{'title': 'Launch Week Recap—make the web AI-ready',
    'url': 'https://www.agentql.com/blog/2024-launch-week-recap',
    'date': 'Nov 18, 2024',
    'author': 'Rachel-Lee Nabors'},
   {'title': 'Accurate data extraction from PDFs and images with AgentQL',
    'url': 'https://www.agentql.com/blog/accurate-data-extraction-pdfs-images',
    'date': 'Feb 1, 2025',
    'author': 'Rachel-Lee Nabors'},
   {'title': 'Introducing Scheduled Scraping Workflows',
    'url': 'https://www.agentql.com/blog/scheduling',
    'date': 'Dec 2, 2024',
    'author': 'Rachel-Lee Nabors'},
   {'title': 'Updates to Our Pricing Model',
    'url': 'https://www.agentql.com/blog/2024-pricing-update',
    'date': 'Nov 19, 2024',
    'author': 'Rachel-Lee Nabors'},
   {'title': 'Get data from any page: AgentQL’s REST API Endpoint—Launch week day 5',
    'url': 'https://www.agentql.com/blog/data-rest-api',
    'date': 'Nov 15, 2024',
    'author': 'Rachel-Lee Nabors'}]},
 'metadata': {'request_id': '0dc1f89c-1b6a-46fe-8089-6cd0f082f094',
  'generated_query': None,
  'screenshot': None}}

ExtractWebDataBrowserTool

• query: 要执行的 AgentQL 查询。如果您想提取精确的结构化数据，请使用 AgentQL 查询。了解如何在文档中 编写 AgentQL 查询，或在 AgentQL 游乐场 中测试一个。
• prompt: 要从页面提取的数据的自然语言描述。AgentQL 将根据您的提示词推断数据结构。如果您想提取由自由格式语言定义的数据而不定义特定结构，请使用 prompt。

注意： 您必须定义 query 或 prompt 才能使用 AgentQL。

要提取数据，首先您必须使用 LangChain 的 Playwright 工具导航到网页。

from langchain_community.tools.playwright import NavigateTool

navigate_tool = NavigateTool(async_browser=async_browser)
await navigate_tool.ainvoke({"url": "https://www.agentql.com/blog"})

'Navigating to https://www.agentql.com/blog returned status code 200'

# You can invoke the tool with either a query or a prompt

# await extract_web_data_browser_tool.ainvoke(
#     {'query': '{ blogs[] { title url date author } }'}
# )

await extract_web_data_browser_tool.ainvoke(
    {"prompt": "the blog posts with title, url, date of post and author"}
)

/usr/local/lib/python3.11/dist-packages/agentql/_core/_utils.py:167: UserWarning: 🚨 The function get_data_by_prompt_experimental is experimental and may not work as expected 🚨
  warnings.warn(

{'blog_posts': [{'title': 'Launch Week Recap—make the web AI-ready',
   'url': 'https://www.agentql.com/blog/2024-launch-week-recap',
   'date': 'Nov 18, 2024',
   'author': 'Rachel-Lee Nabors'},
  {'title': 'Accurate data extraction from PDFs and images with AgentQL',
   'url': 'https://www.agentql.com/blog/accurate-data-extraction-pdfs-images',
   'date': 'Feb 1, 2025',
   'author': 'Rachel-Lee Nabors'},
  {'title': 'Introducing Scheduled Scraping Workflows',
   'url': 'https://www.agentql.com/blog/scheduling',
   'date': 'Dec 2, 2024',
   'author': 'Rachel-Lee Nabors'},
  {'title': 'Updates to Our Pricing Model',
   'url': 'https://www.agentql.com/blog/2024-pricing-update',
   'date': 'Nov 19, 2024',
   'author': 'Rachel-Lee Nabors'},
  {'title': 'Get data from any page: AgentQL’s REST API Endpoint—Launch week day 5',
   'url': 'https://www.agentql.com/blog/data-rest-api',
   'date': 'Nov 15, 2024',
   'author': 'Rachel-Lee Nabors'}]}

GetWebElementBrowserTool

• prompt: 要在页面上查找的 Web 元素的自然语言描述。

selector = await extract_web_element_tool.ainvoke({"prompt": "Next page button"})
selector

"[tf623_id='194']"

from langchain_community.tools.playwright import ClickTool

# Disabling 'visible_only' will allow us to click on elements that are not visible on the page
await ClickTool(async_browser=async_browser, visible_only=False).ainvoke(
    {"selector": selector}
)

"Clicked element '[tf623_id='194']'"

from langchain_community.tools.playwright import CurrentWebPageTool

await CurrentWebPageTool(async_browser=async_browser).ainvoke({})

'https://www.agentql.com/blog/page/2'

链式调用

您可以通过先将工具绑定到 工具调用模型 然后调用它来在链中使用 AgentQL 工具：

实例化 LLM

import os

os.environ["OPENAI_API_KEY"] = "YOUR_OPENAI_API_KEY"

from langchain.chat_models import init_chat_model

model = init_chat_model(model="gpt-4.1", model_provider="openai")

执行工具链

from langchain_core.prompts import ChatPromptTemplate
from langchain_core.runnables import RunnableConfig, chain

prompt = ChatPromptTemplate(
    [
        ("system", "You are a helpful assistant in extracting data from website."),
        ("human", "{user_input}"),
        ("placeholder", "{messages}"),
    ]
)

# specifying tool_choice will force the model to call this tool.
model_with_tools = model.bind_tools(
    [extract_web_data_tool], tool_choice="extract_web_data_with_rest_api"
)

model_chain = prompt | model_with_tools


@chain
def tool_chain(user_input: str, config: RunnableConfig):
    input_ = {"user_input": user_input}
    ai_msg = model_chain.invoke(input_, config=config)
    tool_msgs = extract_web_data_tool.batch(ai_msg.tool_calls, config=config)
    return {"messages": tool_msgs}


tool_chain.invoke(
    "Extract data from https://www.agentql.com/blog using the following agentql query: { posts[] { title url date author } }"
)

{'messages': [ToolMessage(content='{"data": {"posts": [{"title": "Launch Week Recap—make the web AI-ready", "url": "https://www.agentql.com/blog/2024-launch-week-recap", "date": "Nov 18, 2024", "author": "Rachel-Lee Nabors"}, {"title": "Accurate data extraction from PDFs and images with AgentQL", "url": "https://www.agentql.com/blog/accurate-data-extraction-pdfs-images", "date": "Feb 1, 2025", "author": "Rachel-Lee Nabors"}, {"title": "Introducing Scheduled Scraping Workflows", "url": "https://www.agentql.com/blog/scheduling", "date": "Dec 2, 2024", "author": "Rachel-Lee Nabors"}, {"title": "Updates to Our Pricing Model", "url": "https://www.agentql.com/blog/2024-pricing-update", "date": "Nov 19, 2024", "author": "Rachel-Lee Nabors"}, {"title": "Get data from any page: AgentQL’s REST API Endpoint—Launch week day 5", "url": "https://www.agentql.com/blog/data-rest-api", "date": "Nov 15, 2024", "author": "Rachel-Lee Nabors"}]}, "metadata": {"request_id": "1a84ed12-d02a-497d-b09d-21fe49342fa3", "generated_query": null, "screenshot": null}}', name='extract_web_data_with_rest_api', tool_call_id='call_z4Rl1MpjJZNcbLlq1OCneoMF')]}

在代理中使用

您可以在 AI Agent 中使用 AgentQL 工具，使用 AgentQLBrowserToolkit。此工具包包含 ExtractDataBrowserTool 和 GetWebElementBrowserTool。这是一个结合 AgentQL 工具包与 Playwright 工具的代理浏览器操作的示例。

实例化工具包

from langchain_agentql.utils import create_async_playwright_browser

async_agent_browser = await create_async_playwright_browser()

from langchain_agentql import AgentQLBrowserToolkit

agentql_toolkit = AgentQLBrowserToolkit(async_browser=async_agent_browser)
agentql_toolkit.get_tools()

[ExtractWebDataBrowserTool(async_browser=<Browser type=<BrowserType name=chromium executable_path=/root/.cache/ms-playwright/chromium-1155/chrome-linux/chrome> version=133.0.6943.16>),
 GetWebElementBrowserTool(async_browser=<Browser type=<BrowserType name=chromium executable_path=/root/.cache/ms-playwright/chromium-1155/chrome-linux/chrome> version=133.0.6943.16>)]

from langchain_community.tools.playwright import ClickTool, NavigateTool

# we hand pick the following tools to allow more precise agentic browser actions
playwright_toolkit = [
    NavigateTool(async_browser=async_agent_browser),
    ClickTool(async_browser=async_agent_browser, visible_only=False),
]
playwright_toolkit

[NavigateTool(async_browser=<Browser type=<BrowserType name=chromium executable_path=/root/.cache/ms-playwright/chromium-1155/chrome-linux/chrome> version=133.0.6943.16>),
 ClickTool(async_browser=<Browser type=<BrowserType name=chromium executable_path=/root/.cache/ms-playwright/chromium-1155/chrome-linux/chrome> version=133.0.6943.16>, visible_only=False)]

与 ReAct Agent 配合使用

pip install --quiet -U langgraph

from langchain.agents import create_agent


# You need to set up an llm, please refer to the chaining section
agent_executor = create_agent(
    model, agentql_toolkit.get_tools() + playwright_toolkit
)

prompt = """
Navigate to https://news.ycombinator.com/,
extract the news titles on the current page,
show the current page url,
find the button on the webpage that direct to the next page,
click on the button,
show the current page url,
extract the news title on the current page
extract the news titles that mention "AI" from the two pages.
"""

events = agent_executor.astream(
    {"messages": [("user", prompt)]},
    stream_mode="values",
)
async for event in events:
    event["messages"][-1].pretty_print()

================================ Human Message =================================


Navigate to https://news.ycombinator.com/,
extract the news titles on the current page,
show the current page url,
find the button on the webpage that direct to the next page,
click on the button,
show the current page url,
extract the news title on the current page
extract the news titles that mention "AI" from the two pages.

================================== Ai Message ==================================
Tool Calls:
  navigate_browser (call_3eY5a0BRwyYj7kaNpAxkquTD)
 Call ID: call_3eY5a0BRwyYj7kaNpAxkquTD
  Args:
    url: https://news.ycombinator.com/
================================= Tool Message =================================
Name: navigate_browser

Navigating to https://news.ycombinator.com/ returned status code 200
================================== Ai Message ==================================
Tool Calls:
  extract_web_data_from_browser (call_WvRrZKGGo8mq3JewRlaIS5xx)
 Call ID: call_WvRrZKGGo8mq3JewRlaIS5xx
  Args:
    prompt: Extract all the news titles from this page.

/usr/local/lib/python3.11/dist-packages/agentql/_core/_utils.py:167: UserWarning: 🚨 The function get_data_by_prompt_experimental is experimental and may not work as expected 🚨
  warnings.warn(

================================= Tool Message =================================
Name: extract_web_data_from_browser

{"news_item": [{"title": "I Went to SQL Injection Court"}, {"title": "Framework's first desktop is a strange-but unique-mini ITX gaming PC"}, {"title": "Hyperspace"}, {"title": "The XB-70 (2019)"}, {"title": "How core Git developers configure Git"}, {"title": "Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs [pdf]"}, {"title": "Hard problems that reduce to document ranking"}, {"title": "Ggwave: Tiny Data-over-Sound Library"}, {"title": "Bald eagles are thriving again after near extinction"}, {"title": "Forum with 2.6M posts being deleted due to UK Online Safety Act"}, {"title": "Launch HN: Browser Use (YC W25) - open-source web agents"}, {"title": "Part two of Grant Sanderson's video with Terry Tao on the cosmic distance ladder"}, {"title": "New maps of the chaotic space-time inside black holes"}, {"title": "Knitting Your Parachute"}, {"title": "Chicory: A JVM native WebAssembly runtime"}, {"title": "Low Overhead Allocation Sampling with VMProf in PyPy's GC"}, {"title": "Sigma BF Camera"}, {"title": "DeepSearcher: A local open-source Deep Research"}, {"title": "Xonsh - A Python-powered shell"}, {"title": "A possible future of Python in the browser"}, {"title": "Show HN: GoatDB - A lightweight, offline-first, realtime NoDB for Deno and React"}, {"title": "Embedding Python in Elixir, it's fine"}, {"title": "The Deep Research problem"}, {"title": "Why are QR Codes with capital letters smaller than QR codes with lower case?"}, {"title": "Show HN: My new wiki for Silicon Graphics stuff"}, {"title": "AI is blurring the line between PMs and engineers?"}, {"title": "I recreated Shazam's algorithm with Go [video]"}, {"title": "Dogs may have domesticated themselves because they liked snacks, model suggests"}, {"title": "Show HN: Txtl - Fast static website of text utilities"}, {"title": "Have we been wrong about why Mars is red?"}]}
================================== Ai Message ==================================
Tool Calls:
  get_web_element_from_browser (call_B6jn5ItasceNW7eeb640UhQQ)
 Call ID: call_B6jn5ItasceNW7eeb640UhQQ
  Args:
    prompt: button or link to go to the next page
  extract_web_data_from_browser (call_Wyh2VH76bzrlDozp7gpkVBl7)
 Call ID: call_Wyh2VH76bzrlDozp7gpkVBl7
  Args:
    prompt: Extract the current page URL

/usr/local/lib/python3.11/dist-packages/agentql/_core/_utils.py:167: UserWarning: 🚨 The function get_data_by_prompt_experimental is experimental and may not work as expected 🚨
  warnings.warn(

================================= Tool Message =================================
Name: extract_web_data_from_browser

{"current_page_url": "https://news.ycombinator.com/news"}
================================== Ai Message ==================================
Tool Calls:
  click_element (call_NLGIW1lLutkZ6k0vqkfGbOD7)
 Call ID: call_NLGIW1lLutkZ6k0vqkfGbOD7
  Args:
    selector: [tf623_id='944']
================================= Tool Message =================================
Name: click_element

Clicked element '[tf623_id='944']'
================================== Ai Message ==================================
Tool Calls:
  extract_web_data_from_browser (call_QPt8R2hqiSgytUvLcWUUORKF)
 Call ID: call_QPt8R2hqiSgytUvLcWUUORKF
  Args:
    prompt: Extract the current page URL

/usr/local/lib/python3.11/dist-packages/agentql/_core/_utils.py:167: UserWarning: 🚨 The function get_data_by_prompt_experimental is experimental and may not work as expected 🚨
  warnings.warn(

================================= Tool Message =================================
Name: extract_web_data_from_browser

{"current_page_url": "https://news.ycombinator.com/news?p=2"}
================================== Ai Message ==================================
Tool Calls:
  extract_web_data_from_browser (call_ZZOPrIfVaVQ1A26j8EGE913W)
 Call ID: call_ZZOPrIfVaVQ1A26j8EGE913W
  Args:
    prompt: Extract all the news titles from this page.

/usr/local/lib/python3.11/dist-packages/agentql/_core/_utils.py:167: UserWarning: 🚨 The function get_data_by_prompt_experimental is experimental and may not work as expected 🚨
  warnings.warn(

================================= Tool Message =================================
Name: extract_web_data_from_browser

{"news_item": [{"title": "'Hey Number 17 '"}, {"title": "Building and operating a pretty big storage system called S3 (2023)"}, {"title": "Ghost House - software for automatic inbetweens"}, {"title": "Ask HN: Former devs who can't get a job, what did you end up doing for work?"}, {"title": "DeepSeek open source DeepEP - library for MoE training and Inference"}, {"title": "SETI's hard steps and how to resolve them"}, {"title": "A Defense of Weird Research"}, {"title": "DigiCert: Threat of legal action to stifle Bugzilla discourse"}, {"title": "Show HN: Tach - Visualize and untangle your Python codebase"}, {"title": "Ask HN: A retrofitted C dialect?"}, {"title": "“The closer to the train station, the worse the kebab” - a “study”"}, {"title": "Brewing Clean Water: The metal-remediating benefits of tea preparation"}, {"title": "Invoker Commands (Explainer)"}, {"title": "Freelancing: How I found clients, part 1"}, {"title": "Claude 3.7 Sonnet and Claude Code"}, {"title": "Clean Code vs. A Philosophy Of Software Design"}, {"title": "Show HN: While the world builds AI Agents, I'm just building calculators"}, {"title": "History of CAD"}, {"title": "Fans are better than tech at organizing information online (2019)"}, {"title": "Some Programming Language Ideas"}, {"title": "The independent researcher (2018)"}, {"title": "The best way to use text embeddings portably is with Parquet and Polars"}, {"title": "Show HN: Prioritize Anything with Stacks"}, {"title": "Ashby (YC W19) Is Hiring Principal Product Engineers"}, {"title": "GibberLink [AI-AI Communication]"}, {"title": "Show HN: I made a site to tell the time in corporate"}, {"title": "It’s still worth blogging in the age of AI"}, {"title": "What would happen if we didn't use TCP or UDP?"}, {"title": "Closing the “green gap”: energy savings from the math of the landscape function"}, {"title": "Larry Ellison's half-billion-dollar quest to change farming"}]}
================================== Ai Message ==================================

以下是操作和结果的摘要：

### 第 1 页
- **URL:** [https://news.ycombinator.com/news](https://news.ycombinator.com/news)
- **新闻标题：**
  1. I Went to SQL Injection Court
  2. Framework's first desktop is a strange-but unique-mini ITX gaming PC
  3. Hyperspace
  4. The XB-70 (2019)
  5. How core Git developers configure Git
  6. Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs [pdf]
  7. Hard problems that reduce to document ranking
  8. Ggwave: Tiny Data-over-Sound Library
  9. Bald eagles are thriving again after near extinction
  10. Forum with 2.6M posts being deleted due to UK Online Safety Act
  11. Launch HN: Browser Use (YC W25) - open-source web agents
  12. Part two of Grant Sanderson's video with Terry Tao on the cosmic distance ladder
  13. New maps of the chaotic space-time inside black holes
  14. Knitting Your Parachute
  15. Chicory: A JVM native WebAssembly runtime
  16. Low Overhead Allocation Sampling with VMProf in PyPy's GC
  17. Sigma BF Camera
  18. DeepSearcher: A local open-source Deep Research
  19. Xonsh - A Python-powered shell
  20. A possible future of Python in the browser
  21. Show HN: GoatDB - A lightweight, offline-first, realtime NoDB for Deno and React
  22. Embedding Python in Elixir, it's fine
  23. The Deep Research problem
  24. Why are QR Codes with capital letters smaller than QR codes with lower case?
  25. Show HN: My new wiki for Silicon Graphics stuff
  26. **AI is blurring the line between PMs and engineers?**
  27. I recreated Shazam's algorithm with Go [video]
  28. Dogs may have domesticated themselves because they liked snacks, model suggests
  29. Show HN: Txtl - Fast static website of text utilities
  30. Have we been wrong about why Mars is red?

### 第 2 页
- **URL:** [https://news.ycombinator.com/news?p=2](https://news.ycombinator.com/news?p=2)
- **新闻标题：**
  1. 'Hey Number 17'
  2. Building and operating a pretty big storage system called S3 (2023)
  3. Ghost House - software for automatic inbetweens
  4. Ask HN: Former devs who can't get a job, what did you end up doing for work?
  5. DeepSeek open source DeepEP - library for MoE training and Inference
  6. SETI's hard steps and how to resolve them
  7. A Defense of Weird Research
  8. DigiCert: Threat of legal action to stifle Bugzilla discourse
  9. Show HN: Tach - Visualize and untangle your Python codebase
  10. Ask HN: A retrofitted C dialect?
  11. “The closer to the train station, the worse the kebab” - a “study”
  12. Brewing Clean Water: The metal-remediating benefits of tea preparation
  13. Invoker Commands (Explainer)
  14. Freelancing: How I found clients, part 1
  15. Claude 3.7 Sonnet and Claude Code
  16. Clean Code vs. A Philosophy Of Software Design
  17. **Show HN: While the world builds AI Agents, I'm just building calculators**
  18. History of CAD
  19. Fans are better than tech at organizing information online (2019)
  20. Some Programming Language Ideas
  21. The independent researcher (2018)
  22. The best way to use text embeddings portably is with Parquet and Polars
  23. Show HN: Prioritize Anything with Stacks
  24. Ashby (YC W19) Is Hiring Principal Product Engineers
  25. **GibberLink [AI-AI Communication]**
  26. Show HN: I made a site to tell the time in corporate
  27. **It’s still worth blogging in the age of AI**
  28. What would happen if we didn't use TCP or UDP?
  29. Closing the “green gap”: energy savings from the math of the landscape function
  30. Larry Ellison's half-billion-dollar quest to change farming

### 提及“AI”的新闻标题：
1. 第 1 页：**AI is blurring the line between PMs and engineers?**
2. 第 2 页：
   - **Show HN: While the world builds AI Agents, I'm just building calculators**
   - **GibberLink [AI-AI Communication]**
   - **It’s still worth blogging in the age of AI**

---

API 参考

有关如何使用此集成的更多信息，请参阅 git 仓库 或 LangChain 集成文档

---

Edit this page on GitHub or file an issue.

Connect these docs to Claude, VSCode, and more via MCP for real-time answers.