Build an AI News Monitoring Agent with LangChain + wpnews (Python)

In this tutorial, we'll build an autonomous news monitoring agent that watches a list of AI companies, detects when coverage surges, and sends targeted Slack alerts. The agent uses wpnews's get_news_velocity (burst detection) and get_adaptive_context (smart context loading) to minimize API calls while staying responsive to breaking news.

Architecture overview

1. Velocity gate — check /api/v1/news-velocity first. If it's a quiet day, skip the heavy fetch.
2. Watchlist polling — /api/v1/watchlist/digest returns articles for all tracked entities in one call.
3. LLM summarization — use LangChain with Claude to summarize new articles and generate a Slack message.
4. Checkpoint tracking — use /api/v1/articles/since?ts=… to only fetch articles newer than the last check.

Setup

pip install langchain langchain-anthropic requests slack-sdk

import os
WPNEWS_KEY = os.environ["WPNEWS_API_KEY"]
ANTHROPIC_KEY = os.environ["ANTHROPIC_API_KEY"]
SLACK_TOKEN = os.environ["SLACK_BOT_TOKEN"]
SLACK_CHANNEL = "#ai-news-alerts"
WPNEWS_BASE = "https://api.wpnews.pro"
HEADERS = {"X-API-Key": WPNEWS_KEY}

Step 1 — Add companies to your watchlist

Do this once (or via the dashboard at wpnews.pro/dashboard):

import requests

entities = ["OpenAI", "Anthropic", "Google DeepMind", "Mistral AI", "xAI"]
for entity in entities:
    requests.post(
        f"{WPNEWS_BASE}/api/v1/watchlist/items",
        headers=HEADERS,
        json={"type": "entity", "value": entity, "label": entity},
    )

Step 2 — The monitoring loop

import time, json, requests
from datetime import datetime, timezone
from langchain_anthropic import ChatAnthropic
from langchain.schema import HumanMessage
from slack_sdk import WebClient

slack = WebClient(token=SLACK_TOKEN)
llm = ChatAnthropic(model="claude-haiku-4-5-20251001", api_key=ANTHROPIC_KEY)

last_check_ts = datetime.now(timezone.utc).isoformat()

def check_and_alert():
    global last_check_ts

    # Gate 1: velocity check — skip heavy fetch on quiet days
    vel = requests.get(f"{WPNEWS_BASE}/api/v1/news-velocity", headers=HEADERS).json()
    if vel.get("is_quiet"):
        print(f"Quiet day ({vel['velocity_ratio']:.1f}x baseline) — skipping")
        return

    # Gate 2: only new articles since last check
    digest = requests.get(
        f"{WPNEWS_BASE}/api/v1/watchlist/digest",
        headers=HEADERS,
        params={"hours": 1, "limit_per_item": 3},
    ).json()

    new_articles = [
        art for group in digest.get("items", [])
        for art in group.get("articles", [])
        if art.get("published_at", "") > last_check_ts
    ]

    if not new_articles:
        return

    # LLM summary
    article_text = "\n\n".join(
        f"**{a['title']}** ({a.get('entity','')}) — {a.get('summary','')}"
        for a in new_articles[:5]
    )
    response = llm.invoke([HumanMessage(
        content=f"Summarize these AI news items in 2-3 sentences for a Slack alert:\n\n{article_text}"
    )])
    summary = response.content

    # Send to Slack
    burst_prefix = " ⚡ BURST" if vel.get("is_burst") else ""
    slack.chat_postMessage(
        channel=SLACK_CHANNEL,
        text=f"*AI News Update{burst_prefix}* — {len(new_articles)} new articles\n{summary}",
    )
    last_check_ts = datetime.now(timezone.utc).isoformat()
    print(f"Alert sent: {len(new_articles)} articles")

# Run every 15 minutes
while True:
    try:
        check_and_alert()
    except Exception as e:
        print(f"Error: {e}")
    time.sleep(900)  # 15 minutes

Key optimization: adaptive context

When the agent needs to answer questions about the news (not just alert), use adaptive context to avoid over-fetching:

ctx = requests.get(
    f"{WPNEWS_BASE}/api/v1/adaptive-context",
    headers=HEADERS
).json()

# ctx['mode'] is 'quiet' | 'normal' | 'surge'
# ctx['context_text'] is ready for system prompt injection
# ctx['tokens_approx'] tells you the budget used (50-400)

from langchain_anthropic import ChatAnthropic
from langchain.schema import SystemMessage, HumanMessage

llm = ChatAnthropic(model="claude-sonnet-4-6", api_key=ANTHROPIC_KEY)
response = llm.invoke([
    SystemMessage(content=f"You are an AI news analyst. Context: {ctx['context_text']}"),
    HumanMessage(content="What should our ML team pay attention to today?"),
])

Production tips

• Use get_articles/since?ts=… instead of hours-based filtering for checkpoint-precise polling
• Rate limit awareness: wpnews returns X-RateLimit-Remaining headers — check them in your loop
• Free tier (1k/day): with 15-min polling + velocity gate, you'll use ~8-12 calls/hour = ~200/day — well within free tier
• Pro webhooks: instead of polling, use POST /api/v1/webhooks to get push notifications when new articles appear (Pro tier)