langsearch_skill

Overview

This skill provides a free web search and semantic reranking API designed for AGI and LLM applications. It combines keyword search with vector-based semantic matching to retrieve up-to-date web information and improve result relevance. Use it to feed current web context into retrieval-augmented generation (RAG) pipelines and other LLM workflows.

How this skill works

The skill calls LangSearch endpoints to perform hybrid web searches (keyword + vector matching) and optionally returns markdown summaries and relevance scores. It also offers a reranking endpoint that accepts candidate documents and reorders them by semantic relevance to a user query. Typical usage: fetch initial web results, extract snippets or summaries, then rerank candidates before passing context to an LLM.

When to use it

• When your model needs current, real-world information beyond its training cutoff
• To build RAG systems that merge web context with generative LLM responses
• When basic keyword search returns noisy results and you need semantic relevance
• For multi-query research workflows that aggregate and synthesize web findings
• When you need lightweight, production-ready search with summaries and scores

Best practices

• Store LANGSEARCH_API_KEY in environment variables and never commit it to version control
• Request summaries only when needed to reduce response size and latency
• Use freshness filters to limit results to recent time ranges for time-sensitive queries
• Combine initial web-search hits with reranking to filter noise before LLM consumption
• Implement exponential backoff on 429/5xx responses and monitor usage on the dashboard

Example use cases

• Answering user questions with citations by searching the web and passing summaries to an LLM
• Running multi-topic research batches and aggregating top results for analyst review
• Reranking internal search results or scraped documents to surface the most semantically relevant items
• Enriching chatbots with current news, policies, or product updates in real time
• Building lightweight discovery tools that return concise summaries and relevance scores