What is a Grounding API?
Large language models answer from their training data — which is frozen in time and can hallucinate plausible-sounding facts. A Grounding API connects the model to a live information source (web search, a curated index, or your own documents) so it can cite real sources, stay current on recent events, and admit when it doesn’t know.
For AI agents this matters even more: an agent that decides which tool to call, what email to send, or what code to ship needs facts it can trust. Grounding is the difference between “the CEO is probably X” and “the CEO is Y — here’s the source”.
This dashboard:
- Fans the same question out to multiple grounding-capable providers at once.
- Runs each model twice — grounded (with web/search tools) and a baseline (no tools) — so you can see the delta.
- Uses an LLM-as-judge to score every response 1–5 on correctness, groundedness, and citation support.
- Reports latency, tokens, cost, and the exact citations each provider returned.
Try a freshness-sensitive question from the buttons below — the gap between grounded and baseline is usually dramatic.
How this dashboard works
Providers: