Skip to content
webxwebx

Search & answer commands

Section titled “search”

webx search <q> — metasearch over 15 providers — ddg, hn, so, wiki, index, gh, reddit, sg, npm, crates, searxng, brave, mojeek, kagi, grep — fused by weighted RRF + BM25 blending + domain priors. score is normalized 0–1; near-duplicates drop via title + simhash (deduped count in output).

Filters

Section titled “Filters”

--site / --domains / --exclude-domains · --after / --before · --lang · --topic · --exact · --location <cc> (geo hint — Exa userLocation/Firecrawl location parity) · --category developer (restricts providers to code/docs sources: so, gh, grep, sg, npm, crates, hn, reddit)

Depth & sources

Section titled “Depth & sources”
Terminal window
webx search <q> --depth advanced              # Tavily-style tier: fast|basic|advanced
webx search <q> --sources web,news,images     # searxng category verticals
webx search <q> --subpages 3 --subpage-target a,b   # read matching subpages per result (Exa-style)
Section titled “Read-after-search”
Terminal window
webx search <q> --scrape                      # fetch top results inline (bounded, per-page timeouts)
webx search <q> --scrape --highlights         # keep only query-relevant excerpts
webx search <q> --rerank                      # snippet-level rerank, no page fetches
webx search <q> --scrape --rerank             # content-level rerank on the scraped pages
webx search <q> --answer                      # attach a cited extractive answer (Tavily include_answer)
webx search <q> --fresh                       # bypass the 10m result cache

Rerank uses embedding cosine when WEBX_EMBED_MODEL is configured and a term-coverage lexical score otherwise — opt-in, so search stays fast without it. depth:advanced reranks automatically.

ask

Section titled “ask”

webx ask <q> — search → scrape → citation-numbered excerpts under a token budget. One shot, source-backed:

Terminal window
webx ask "how does htmx hx-swap work"
webx ask "go generics constraints" --llm      # LLM-synthesized cited answer (WEBX_LLM_*)

research

Section titled “research”

webx research <question> — deep-research pipeline: expand → search → read sources → cited markdown report.

Terminal window
webx research "state of wasm in 2026" --sources 8
webx research <q> --schema '{"title":"string","findings":["string"]}'   # JSON matching schema (Tavily output_schema equivalent)

LLM-synthesized when WEBX_LLM_* is configured; falls back to ranked excerpts otherwise.

verify

Section titled “verify”

webx verify <claim> — fact-check against live web sources → supported | refuted | unclear + confidence + cited sources. The grounding check (Jina g.jina.ai equivalent); honestly reports unavailable without an LLM.

eval

Section titled “eval”

webx eval — score search quality on the bundled 30-query dev set → hit rate + MRR, per-kind breakdown. --extract scores content extraction quality. eval --vs exa,tavily,firecrawl re-scores the same corpus through each engine when you have their keys — the honest-benchmark harness.

doctor

Section titled “doctor”

webx doctor — probe every provider: liveness, latency, result counts, config errors. Run this first when results look thin.