Tavily Web Search¶
Performs a live web search using the Tavily search provider and returns a concise answer (optional) along with structured search results. The node calls the Agents service and outputs a JSON-formatted string of documents and, when requested, a short synthesized answer.
Usage¶
Use this node when you need up-to-date information or sources from the web to ground an answer or collect references. Typical workflows: answer a user question and cite sources, gather recent news on a topic, or filter results to/from specific domains. Configure depth, topic, and domain filters to control the scope and recency of results.
Inputs¶
| Field | Required | Type | Description | Example |
|---|---|---|---|---|
| query | True | STRING | The search query text to send to Tavily. | What are the latest breakthroughs in quantum error correction? |
| search_depth | True | CHOICE | Depth of the search. Use 'basic' for a quicker, lighter search or 'advanced' for a deeper search. | advanced |
| max_results | True | INT | Maximum number of search results to return. | 5 |
| topic | True | CHOICE | Category of the search. Choose 'general' for broad results or 'news' to prioritize recent news. | news |
| days | True | INT | Only used when topic is 'news'. Number of days back from today to include in results. | 3 |
| include_answer | True | BOOLEAN | Whether to include a short synthesized answer to the query. | true |
| include_raw_content | True | BOOLEAN | If true, include raw text content from results; otherwise return only the most relevant extracted content. | false |
| include_domains | True | STRING | Limit results to these domains. Provide one domain per line. | arxiv.org nature.com |
| exclude_domains | True | STRING | Exclude results from these domains. Provide one domain per line. | wikipedia.org reddit.com |
Outputs¶
| Field | Type | Description | Example |
|---|---|---|---|
| answer | STRING | A short, synthesized answer to the query if requested. | Quantum error correction progressed with bosonic codes enabling lower overhead fault tolerance; see sources for details. |
| documents | STRING | A JSON-formatted string containing the list of retrieved documents, including titles, URLs, snippets, and optionally raw content. | [{"title": "Paper on bosonic codes", "url": "https://arxiv.org/abs/xxxx.xxxxx", "snippet": "Recent advances in..."}] |
Important Notes¶
- Query must be a non-empty string; empty queries will cause an error.
- If 'include_answer' is disabled, the 'answer' output may be empty or omitted by the service; rely on 'documents' for sources.
- The 'days' parameter is applicable only when 'topic' is set to 'news'; otherwise it is ignored.
- Setting 'include_raw_content' to true can increase response size and processing time.
- Search depth affects the breadth/comprehensiveness of results; advanced depth may be slower or more resource-intensive.
- Domain filters expect one domain per line; avoid protocols (http/https) and paths.
Troubleshooting¶
- Empty or whitespace-only 'query' causes an immediate validation error. Provide a meaningful query.
- If the node raises a request failure error, verify the Agents service is reachable and try again later.
- If the response parsing fails, reduce 'include_raw_content' or 'max_results' to lower payload size and retry.
- No answer returned: ensure 'include_answer' is set to true; otherwise use the 'documents' output to compose an answer.
- Unexpected or irrelevant results: refine 'query', switch 'search_depth' to 'advanced', or adjust domain include/exclude lists.
- Old news in results: set 'topic' to 'news' and tune 'days' to the desired recency window.
Example Pipelines¶
