Agent skills for integrating Valyu APIs into AI agents and assistants.
These skills follow the Agent Skills specification, making them compatible with Claude Code, Claude Desktop, and other agent platforms.
Valyu provides a unified API for AI agents to access real-time information across:
- Web - General internet content
- Academic - arXiv, PubMed, bioRxiv, medRxiv (40M+ papers)
- Financial - SEC filings, earnings, stock data, crypto
- Healthcare - Clinical trials, drug labels, ChEMBL, DrugBank
- Economic - FRED, BLS, World Bank data
- News - Real-time news with date/country filtering
- Predictions - Polymarket, Kalshi odds
- Transportation - UK Rail, ship tracking
- Patents - US patent database
| Skill | Description |
|---|---|
| valyu-search | Complete Valyu API toolkit - search, content extraction, AI answers, deep research |
npx skills add valyuAI/skillsThe skill provides instructions for:
- Search API - Find information across 25+ data sources
- Contents API - Extract clean markdown from URLs
- Answer API - Get AI-synthesized answers with citations
- DeepResearch API - Generate comprehensive research reports
-
Get a Valyu API key at platform.valyu.ai ($10 free credits)
-
Install the SDK:
npm install valyu-js # or pip install valyu -
Use the API:
import Valyu from 'valyu-js'; const valyu = new Valyu("your-api-key"); // Search across all sources const results = await valyu.search({ query: "transformer architecture attention mechanism 2024", searchType: "all", maxNumResults: 10 }); // Get AI-powered answer const answer = await valyu.answer({ query: "What are the latest developments in quantum computing?" }); // Extract content from URL const content = await valyu.contents({ urls: ["https://example.com/article"], summary: "Extract key findings in 3 bullet points" }); // Deep research (async) const task = await valyu.deepResearch.create({ query: "AI chip market competitive landscape 2024", model: "standard" // fast (~5min), standard (~15min), heavy (~90min) });
valyu-agent-skills/
├── README.md
└── valyu-search/
├── scripts/ # CLI tools
└── best-practices/
├── SKILL.md # Main skill file (start here)
└── references/
├── api-guide.md # Complete API documentation
├── recipes.md # Index of all 27 recipes
├── datasources.md # Available data sources
├── prompting.md # Query writing best practices
├── design-philosophy.md
├── search-recipes/ # 14 search patterns
├── content-recipes/ # 6 content extraction patterns
├── answer-recipes/ # 4 answer generation patterns
├── deepresearch-recipes/ # 3 research depth levels
└── integrations/ # 12 platform guides
What do you need?
├─ Find information across multiple sources
│ └─ Search API (/v1/search)
│
├─ Extract content from specific URLs
│ └─ Contents API (/v1/contents)
│
├─ Get an AI-synthesized answer with citations
│ └─ Answer API (/v1/answer)
│
├─ Generate a comprehensive research report
│ └─ DeepResearch API (/v1/deepresearch)
│
└─ Discover available data sources
└─ Datasources API (/v1/datasources)
| Recipe | Use Case |
|---|---|
| basic-search-all | Search across all sources |
| academic-search | Research papers (arXiv, PubMed) |
| finance-search | SEC filings, earnings, stocks |
| news-search | Real-time news with filtering |
| healthcare-and-bio-search | Clinical trials, drug data |
| Recipe | Use Case |
|---|---|
| basic-content-extraction | Clean markdown from URLs |
| structured-extraction | Extract data via JSON schema |
| Recipe | Use Case |
|---|---|
| basic-answer | AI answers with citations |
| answer-with-streaming | Progressive response generation |
| Recipe | Duration | Use Case |
|---|---|---|
| fast-research | ~5 min | Quick lookups |
| standard-research | ~15 min | Standard research |
| heavy-research | ~90 min | Comprehensive analysis |
| Platform | Guide |
|---|---|
| Anthropic Claude | integrations/anthropic.md |
| OpenAI | integrations/openai.md |
| Vercel AI SDK | integrations/vercel-ai-sdk.md |
| LangChain | integrations/langchain.md |
| LlamaIndex | integrations/llamaindex.md |
| MCP Server | integrations/mcp-server.md |
Keep queries under 400 characters. Be specific, not verbose.
BAD: "I want to know about AI"
GOOD: "transformer attention mechanism survey 2024"
BAD: "Apple financial information"
GOOD: "Apple revenue growth Q4 2024 earnings SEC filing"
Split complex requests into multiple queries:
# Don't do this
"Tesla stock, products, and news"
# Do this instead
Query 1: "Tesla stock performance Q4 2024"
Query 2: "Tesla Cybertruck production updates"
Query 3: "Tesla FSD autonomous driving progress"
See prompting.md for the complete guide.
- Valyu Platform - Get API key
- Valyu Documentation - Official docs
- Agent Skills Specification - Skill format spec
- valyu-js - TypeScript/JavaScript SDK
- valyu (Python) - Python SDK
MIT