Web Search, Page Scraping & Real-Time News
Live real-time search, webpage extraction and scraping, and real-time news retrieval with source citations and zero search history profiling.
Real-Time Information Retrieval
AI models have training cutoff dates. When you need current financial figures, today's news headlines, recent software releases, or up-to-date documentation, relying on a static model risks hallucinations or outdated answers.
ARMES equips your agents with built-in Live Web Search, Webpage Scraping & Extraction, and Real-Time News retrieval tools.
Plan Availability
Live Web Search, Webpage Scraping, and Real-Time News retrieval are available to all users across every plan (Free, Eco, Pro, and Ultra).
How Live Web Search Works
When you ask a question requiring live knowledge:
User: "What were the major announcements at the AI conference this week?"- Query Formulation: The agent recognizes the temporal requirement and crafts concise 3-to-6-word search queries with explicit recency anchors (e.g., "latest", "recent", "2026").
- Sub-Second Live Retrieval: ARMES queries live web indexes, fetching ranked sources, token-dense page excerpts, publication dates, and authoritative domain indicators.
- Source Diversity & Filtering: Results are automatically filtered to strip repetitive navigation boilerplate and capped per root domain, ensuring answers draw from diverse perspectives.
- Grounded Synthesis: The agent reads the retrieved content and synthesizes a concise, grounded response with verified inline citation links
([Source Title](url)). - Execution Safety Caps: To prevent runaway latency or infinite loops, live search executions are capped at a maximum of 6 queries per conversational turn.
Webpage Content Extraction & Document Reading
Beyond searching for keywords, agents can read and analyze specific web links and online documents:
User: "Read https://docs.example.com/api/v2 and summarize the authentication requirements."Key Capabilities:
- Direct Link Extraction: When you paste a URL directly into your prompt, your agent extracts clean Markdown text to inspect articles, API documentation, changelogs, and technical specs.
- Autonomous Follow-Up: When an initial web search uncovers an authoritative article or whitepaper, the agent can autonomously read the full page to gather deep specifics before answering.
- Online PDF Reading: Agents can parse and extract text from publicly accessible online PDFs and research papers.
- Resilient Fallback: If a client-rendered single-page app (SPA) or paywall blocks complete text extraction, the agent automatically falls back to targeted keyword search to retrieve indexed excerpts without hallucinating.
Dedicated Real-Time News & Recency Grounding
When you ask about current events, geopolitical developments, breaking updates, or market movements, the agent automatically applies Temporal Recency Grounding:
- Temporal Detection: Queries containing terms like "breaking", "latest", "today", "yesterday", "this week", or "recent update" trigger strict recency policies that filter for newly published articles.
- Multi-Source Balanced Attribution: The agent aggregates coverage across multiple reputable outlets to balance editorial perspectives and present an objective consensus.
- Epistemic Hygiene: On fast-breaking stories where details are fluid, agents clearly distinguish confirmed facts from early reports and present conflicting claims with neutral attribution.
The Three Tiers of Web Retrieval in ARMES
ARMES provides three distinct levels of web retrieval depending on the complexity of your inquiry:
For complex competitive analyses, academic investigations, or comprehensive market reports, see ARMES Research & Deep Analysis.
Privacy Invariants for Web Search
Search Privacy Guarantee
Unlike consumer search engines that tie search queries to your personal IP address, personal accounts, or advertising profiles, ARMES executes all search and scraping requests through isolated server-side proxy infrastructure. Upstream search providers never receive your user ID, email address, IP address, or device fingerprints.