tavily search

Web search returning LLM-optimized results with content snippets and relevance scores.

Before running any command

Check tvly --status. If tvly is not on PATH:

pip install -q tavily-cli || uv tool install tavily-cli

Auth comes from the TAVILY_API_KEY environment variable (already set in this environment). Never print it. If it is unset, report that and fall back to the built-in web tools.

Quick start

# Basic search
tvly search "your query" --json

# Precision search with more results
tvly search "trellis coded quantization LLM" --depth advanced --max-results 10 --json

# Recent items only
tvly search "open-weight model release" --time-range week --topic news --json

# Primary sources only
tvly search "KV cache quantization" --include-domains arxiv.org,github.com --json

# Full page content inline (saves a separate extract call)
tvly search "vLLM disaggregated serving" --include-raw-content --max-results 3 --json

Key options

Tips

• Keep queries under 400 characters; break complex questions into sub-queries.
• --include-domains arxiv.org,github.com,huggingface.co biases toward primary sources.
• Search snippets are NEVER citable evidence on their own — open the URL with

the tavily-extract skill first (see radar-source-verify).