When a conversation requires current information — recent events, live prices, new research — Tripplet can search the web. The feature is available in all plans. Here's exactly how it works.
When search activates
Search activation is automatic when the model detects that a question likely requires current information. Signals include: temporal language ("latest", "today", "this week"), queries about people or companies, and factual questions where the model's training data might be stale.
You can also trigger search manually by clicking the search icon in the input bar. In manual mode, every response will include web results regardless of the question type.
The search pipeline
When search activates, we run a query against our search index (which combines multiple providers) and retrieve the top 10 candidate pages. We then run a relevance pass using Majuli 3.1 to score and filter to the 3-5 most relevant results. These results are chunked and injected into the model's context before generation.
We explicitly prompt the model to cite sources inline — not at the end of the response, but immediately after the claim they support. This makes it easy to verify specific facts without reading an end-of-post list.
Source cards and expandability
Below each response that uses search, we render expandable source cards — one per source cited. Each card shows the domain, title, and a brief excerpt. Clicking the card opens the full source URL in a new tab.
We intentionally surface the sources prominently because we believe AI responses should be verifiable. If Tripplet tells you something based on a web search, you should be able to check the source in one click.