Lonestone is an AI development studio helping fast-growing tech companies ship production-ready AI products. Their latest project, Mom3nt, is a smart assistant built for sales teams — one that plugs into HubSpot, analyzes each prospect and client’s engagement history, and tells reps who to follow up with, when and why.

But Mom3nt goes one step further: it also generates high-quality, personalized icebreakers based on a contact’s interests, activities, and recent news. And to do that reliably, Lonestone needed something the usual LLM stack couldn’t provide: real-time, contextual web search.

The Challenge: Finding Reliable, Up-to-Date Information About Any Prospect

A good icebreaker, a good follow-up  feels personal and timely. A bad one feels generic or flat.

Lonestone wanted Mom3nt to surface fresh articles, blog posts, interviews, and niche updates around any prospect or client’s interests, without relying on manual research or static enrichment tools.

Their constraints:

• They needed real-time, granular web search (not “LLM knowledge,” not cached news feeds).

• They needed full control over where the search happens — including the ability to whitelist domains.

• They needed a system that wouldn’t hallucinate or invent sources.

• It had to run entirely on-demand, inside icebreaker generation workflows.

In short: they needed high-precision retrieval with no noise.

The Solution: Linkup as the Real-Time Retrieval Layer for Mom3nt

Lonestone integrated Linkup directly into their multi-step best next action generation process. Here’s how Mom3nt now builds an icebreaker: 

1. Profile Understanding
   A first LLM analyzes the contact’s LinkedIn profile, posts, and comments to extract interests and topics.

   
   

2. Domain Discovery
   A second LLM proposes a list of reputable websites related to those interests — effectively using the LLM’s training data to bootstrap domain knowledge.

   
   

3. High-Precision Web Search (Linkup)
   Lonestone sends a specialized prompt to Linkup along with the domain whitelist, ensuring the system fetches only relevant, recent, and trustworthy results.

   
   

4. Result Ranking
   A final LLM chooses the best link among the search results.

   
   

5. Icebreaker / Follow-up Generation
   Mom3nt generates an outreach-ready message leveraging that content. A contextualised message with the right tone of voice, the right content on a perfect timing,  in order to build relationships during the sales cycle at scale. A process that would have take hours per day per sales is now automatic and only takes minutes for the sales teams to execute.

This combination — LLMs for semantic understanding + Linkup for real-time factual grounding — allowed Lonestone to drastically improve icebreaker quality.

“The domain-whitelisting step powered by Linkup significantly boosted the relevance of our results. It’s now the most reliable part of our retrieval pipeline.” — Lonestone Team

The Impact: Higher Relevance, Less Manual Research, More Authentic Sales Conversations

Since integrating Linkup, Lonestone reports: 

Cleaner, more relevant search results

Linkup’s domain controls reduced noise and eliminated irrelevant sources, especially for niche professional topics.

Faster icebreaker generation

What previously required trial-and-error now runs in a single, deterministic pipeline.

Better user experience for sales reps

Mom3nt can now present a link that genuinely feels tailored to the prospect,  improving response quality and trust.

More human outreach, not automated spam

Mom3nt never sends messages automatically, Linkup simply helps provide better raw material for the salesperson.

Here’s a real example Mom3nt generated (anonymized):

“Bonjour Ludovic, je suis tombé récemment sur un article qui explore comment partager les coulisses de son agence pour accélérer la croissance… Je me suis dit que ça ferait écho à votre approche du business vlogging.”

This level of contextualization is only possible when the retrieval layer is precise, fresh, and controllable.

What’s Next

Lonestone is now exploring deeper Linkup features,  including multi-source aggregation and structured outputs,  to further enrich Mom3nt’s recommendations.