Saved by resolve() / working agent comparisons · 1,000 free requests a month
Smart agents. Wrong identities.
Your agent’s decisions depend on context.
Same documents. Same entities to investigate. Two ways to establish identity. One consequential decision for each agent. See what changes when an agent uses resolve()—and what remains unknown.
For a fair comparison, both sides investigate these exact spans. Edit them to try another case. Unmentioned words are not counted as NIL.
Choose a demo and run it. Results are fetched live, not prerecorded.
Without resolve()
String-search agent
First Wikidata name-search result, cached for up to 24 hours. No contextual disambiguation.
No run yet.
With resolve()
Identity-aware agent
Same mention spans, live contextual resolution or explicit NIL; then Wikidata context.
No run yet.
What this compares: The baseline is a transparent string-search agent, not a secretly prompted LLM. Both agents use Wikidata entity descriptions and the same deterministic task-routing logic; only the identity-selection step differs. A missing extraction or Wikidata search hit is not NIL. NIL means resolve() explicitly declined a supplied mention. Unresolved does not prove a company or concept is new. False links may occur on either side.
resolve() results are live. Wikidata first-hit searches may be cached for up to 24 hours. This is a deterministic identity-routing comparison, not an LLM benchmark. Errors are shown, not invented.
Run it on your own text. A free key gives you 1,000 requests a month, the full API and your own vocabulary or knowledge graph.