Docs/Inference/Web search

Web search

Ground any answer in live results. One suffix, no extra tooling.

venym/qwen-27b:online
web_search
userWhat changed in the latest PostgreSQL release?
answerA new major version ships every year, and the release notes group every change by area. Moving between majors uses pg_upgrade, or a dump and restore.

Turning it on

Two equivalent ways:

POST /v1/chat/completionscURL
# 1. suffix on the model idcurl https://venym.io/v1/chat/completions \  -H "Authorization: Bearer $VENYM_API_KEY" -H "Content-Type: application/json" \  -d '{"model": "venym/qwen-27b:online",       "messages": [{"role": "user", "content": "What changed in the latest Node.js LTS?"}]}' # 2. venym.web_search in the bodycurl https://venym.io/v1/chat/completions \  -H "Authorization: Bearer $VENYM_API_KEY" -H "Content-Type: application/json" \  -d '{"model": "venym/qwen-27b",       "venym": {"web_search": true, "search_results": 8},       "messages": [{"role": "user", "content": "What changed in the latest Node.js LTS?"}]}'

venym.search_results sets how many results to fetch (default 6, max 10).

What happens

  • We take the last user message as the query (the first 400 characters), run a search, and put the results in a system message ahead of your messages.
  • The model is told the results' titles, URLs and snippets, and to cite them as [n], where n is the result's position.
  • The same list comes back as structured data: venym.sources in the response body, or the leading venym.sources event when streaming.
payload.jsonJSON
"venym": {  "sources": [    { "title": "Node.js 24 release notes", "url": "https://nodejs.org/en/blog/release/..." },    { "title": "Another source", "url": "https://example.com/..." }  ]}

The model's text says things like "as of this release [1]". Map [1] to sources[0].

Cost

A successful search adds $0.0020 to that request, on top of the token cost. The search results also enter the prompt, so you pay input-token price for them. No results, or search failing, means no search fee.

Privacy of the query

The query text goes to our search provider. That is the one place a piece of your prompt leaves for a third party, and only when web search is on. We do not store it. Details on the privacy page.

Search without a model

If you only want results, call the search API directly and skip the model.