Tavily

Agent-first web search JSON

Get factual, source-rich results tailored for LLM agents. Tavily aggregates pages from across the web, filters and ranks them with AI, and returns a compact JSON object with links, snippets, and concise findings. Because the call runs a real browser and de-duplicates noise, your chain sees what users would. Use it for research, grounding, monitoring, or evaluations—without babysitting scrapers, rotating IPs, or parsing brittle HTML. Optionally persist results to compare models and prompts later.

Content extraction & caching

Pull full-text from web pages when you need raw content for indexing or embeddings. The extraction API grabs readable text, stripping boilerplate while preserving headings and links for traceability. Pair this with per-query caching or archives to stabilize datasets across runs, so evaluations, RAG builds, or compliance reviews aren’t skewed by page changes. You get consistent inputs for chunking and retrieval without maintaining crawl code or manual copy-paste workflows that fail at scale.

Control depth, recency & domains

Tune search behavior for your task: choose shallow or exhaustive depth, set time ranges to focus on fresh material, and include or exclude domains to steer quality. Topical filters reduce spammy results while keeping diverse viewpoints for analysis. These controls help agents avoid rabbit holes, keep costs predictable, and produce cited answers fast. Pair with quick summaries for scans, escalating to extraction only on promising links so pipelines stay rigorous and efficient.

SDKs, playground & credits

Start in minutes with the Python SDK and interactive playground to test parameters and view JSON output. Track usage and manage multiple API keys with per-key limits to protect budgets in multi-tenant apps. Generous free credits help teams prototype research, monitoring, and RAG grounding without standing up scrapers. Clear rate-limit guidance and stable endpoints keep CI tests reliable, while webhooks and archives support reproducible runs that make evaluation and governance less painful.

LangChain & CrewAI integrations

Plug Tavily into popular agent frameworks so chains can search with one line. Integrations for LangChain and CrewAI expose parameters like depth, topic filters, time windows, and domain allow/deny lists. Compose search, summarization, and extraction as tools that pass cited context to prompts—reducing glue code and keeping experiments consistent. As needs grow, swap models or prompts while the web layer stays stable and agents keep producing grounded, auditable answers.