Ex Libris Integrates Primo Research Assistant Into Primo VE Discovery
Ex Libris, a Clarivate company, has integrated the Primo Research Assistant into its Primo VE discovery interface, utilizing a Retrieval-Augmented Generation (RAG) architecture powered by OpenAI’s GPT-4.1 Mini model.
The feature, which requires libraries to manually enable it, processes user queries by converting them into Boolean searches, re-ranking results from the Central Discovery Index (CDI), and generating overviews based on descriptive metadata.
The Tech TL;DR:
- The assistant generates summaries strictly from abstracts and descriptive metadata, explicitly excluding full-text analysis, which limits its utility for comprehensive literature reviews.
- The tool is gated by library configuration; it is off by default and requires an active user sign-in, with administrators controlling the inclusion of local institutional records.
- Content exclusions are major—including JSTOR, Elsevier, and DataCite—meaning the AI-generated overview may omit critical sources that would appear in a standard database search.
System Processes Queries Through Five-Step Pipeline
The system operates as a five-step pipeline triggered when a signed-in user submits a natural-language query. According to the official Ex Libris documentation, the model first expands the query into a Boolean string, incorporating variations in both the user’s language and English. This query executes against the CDI, and if configured, the institution’s Alma-managed local inventory.
The system then performs a re-ranking process using embeddings to isolate the top five most relevant sources. The final overview is synthesized by the LLM using only these sources’ abstracts.
Because the tool relies on metadata rather than full-text indexing, the quality of the response is tethered directly to the richness of the library’s cataloging records. To understand how the system constructs these queries, users can trigger a standard search after an AI-assisted query, which reveals the underlying Boolean logic.
# Example: Logical flow for RAG query construction
query_expansion = {
"original": "impact of climate change on agriculture",
"boolean_logic": "(climate change OR global warming) AND (agriculture OR farming OR crops)"
}
# The system maps this to CDI facets and retrieves top 30 candidates for re-ranking
Assistant Excludes Major Publishers from Search Results
The scope of the Primo Research Assistant is bounded by explicit provider-level exclusions. As noted by Ex Libris, the assistant does not access content from DataCite, Elsevier, JSTOR, or Condé Nast. This creates a functional gap where a user might receive an “overview” that lacks primary research from major publishers, despite those items being discoverable through traditional Primo search facets.
The CARLI webinar summary from October 31, 2024, highlighted that these gaps can confuse patrons, as the “5 most relevant sources” label may imply a comprehensive search that the underlying architecture cannot fulfill. During the early rollout, feedback from 26 survey respondents indicated that many users did not feel prepared for the tool’s current limitations, leading to concerns regarding its role in academic research workflows.
| Development Milestone | Event Description |
|---|---|
| June 2024 | Beta testing phase initiated across 18 institutions globally. |
| September 18, 2024 | Official product launch announcement by Clarivate. |
| October 31, 2024 | CARLI webinar documents initial user concerns regarding scope and accessibility. |
For research-intensive environments, the tool functions best as a discovery pointer rather than a definitive authority. Because the model can return sources that are inaccessible to specific patrons, it is recommended that institutions configure their discovery views to prioritize full-text availability. Administrators can manage these settings, including the ability to enable or disable the use of local records, via the Discovery configuration menu.
The reliance on GPT-4.1 Mini necessitates a cautious approach to fact-checking. As documented in the Ex Libris Knowledge Center, the model is prone to the standard risks associated with large language models. The system provides a “View more results” option to revert to a conventional search, which remains the primary requirement for systematic literature reviews that demand high recall and precision.
Ex Libris began as an internal project at the Hebrew University of Jerusalem in 1980, created to manage both Hebrew and Latin character sets. The effort produced ALEPH, which by 1995 was used by 200 libraries in 27 countries, including the seventeen libraries of the Vatican.
In 1995 the commercial arm merged with the university spin-off to form Ex Libris Group. The company launched Primo, the discovery service now hosting the Research Assistant, in May 2007, and released Alma, the first SaaS cloud-based library services platform, in January 2011. ProQuest acquired Ex Libris in December 2015; Clarivate acquired ProQuest in May 2021.
Always consult with certified IT and cybersecurity professionals before altering enterprise networks or handling sensitive data.