Research & knowledge graphs

research — full research in one command

omniscout research "emerging quantum computing startups 2026"
omniscout research "vector database benchmarks 2026" --results 10

One call researches across many sources. Tune it with:

research output pipes cleanly to jq: omniscout research "topic" | jq '.passages[0:3]'.

When to use research vs answer vs extract

• answer — one fact, fast.
• research — a topic you need surveyed across many sources, with passages you can cite.
• extract — you already know the URL(s); skip the search step. See Extraction.

graph — structured knowledge graphs

omniscout graph "Cursor"                 # search the web and extract
omniscout graph "Cursor" -w cursor.com  # use that site directly
omniscout graph "Cursor" --data         # include sources and timing
omniscout graph "Cursor" --llm          # optional AI overlay on the results

graph maps a company, person, or product into a structured tree of entities and relationships. Pass a bare domain (cursor.com) to stay on-site, or -w/--website to use one site and its internal links. --results/-k controls how many sources are used (default 5).

benchmark — compare answer modes

omniscout benchmark answers
omniscout benchmark retrieval
omniscout benchmark ablation   # full ablation matrix with persisted metrics

benchmark compares answer modes against sample questions — useful when deciding between --probe, --classic, and --probe-pro for your workload.

Probe Zero vs Classic

Probe Zero is an answer mode (see omniscout settings engines):

• Probe Zero Mini — good default for direct answers.
• Classic — deterministic, works offline after setup.
• Probe Pro — for the hardest questions.

Compare them with omniscout benchmark answers before standardizing in an agent loop.