“what are all the customers (Companies) and their contacts (if any, their names) of exa.ai and parallel.ai and tavily.com? These customers must be AI labs or agent platforms, preferably those building frontier models and the products on top of them.”
“search x.com and linkedin.com and other less known sources to see if there are additional large/well-known customers of these vendors”
A follow-up sweep of X, LinkedIn, official technical documentation, GitHub-indexed material, podcasts, engineering posts, Reddit, Substack and other lesser-known research sources adds 6 high-confidence public-use relationships to the original 45 — for a total of 51 relationships across 49 unique organizations. Five of the six fit the AI-lab / agent-platform scope strictly; Y Combinator is a prominent adjacent organization with explicit internal agent use. These are publicly evidenced use relationships, not proof of commercial procurement contracts.
Not promoted: Together AI (reported internal GTM research on Exa) and Granola (reportedly relying on Parallel APIs daily) stayed out because the accessible evidence was single-source or vendor-side. Ecosia and Standard Metrics fall outside the large AI-lab / agent-platform focus, and a Harvey-on-Exa lead was a weak third-party mention only.
| Vendor | Company | Contact | Tier / evidence | Use | Source |
|---|
Method: 51 publicly evidenced vendor–customer/use relationships (45 from the original vendor-page shortlist plus 6 from an expanded search of X, LinkedIn, official documentation, GitHub-indexed material, podcasts, Reddit, Substack and research sources), covering 49 unique organizations; compiled November 2025. Each row counts one vendor–company relationship with its strongest public source; evidence badges distinguish direct customer statements, official product documentation, and vendor statements. Blank contacts mean no defensible customer-side contact was isolated, not that none exists. Relationships are public evidence of use, not proof of procurement contracts. Two weaker leads (Together AI, Granola) and out-of-scope finds (Ecosia, Standard Metrics) were cut.