Last week, Cohere introduced Embed 4, a new artificial intelligence (AI) embedding tool aimed at businesses developing and implementing AI applications and agents. Based in Toronto, the company specializes in enterprise-oriented AI models and tools, asserting that Embed 4 can comprehend complex, multimodal documents and effectively retrieve essential information needed for task completion. The tool also promises to reduce data storage expenses by utilizing compressed embeddings rather than full documents.
In a blog update, Cohere provided insights on the newly launched Embed 4, describing it as a multimodal embedding tool that enhances search and retrieval capabilities in current AI systems. The product is available for direct purchase through the company’s website, as well as on platforms like Microsoft Azure AI Foundry and Amazon SageMaker. Moreover, businesses can deploy the tool privately in any virtual private cloud (VPC) or on their own premises.
Embed 4 employs a mechanism known as Retrieval-Augmented Generation (RAG), allowing the AI to extract information from its knowledge base. This system functions by prompting searches based on specific keywords, rankings, and various rule-based algorithms. Essentially, Embed 4 takes over this role for data sourced externally.
The embedding tool is designed to integrate seamlessly with any existing AI infrastructure, whether that is an AI application or an agent. Typically, enterprises utilize either third-party AI models with integrated search engines or develop customized search solutions. Cohere asserts that Embed 4 is a superior alternative to both approaches.
A key feature of Embed 4 is its support for multimodality. The tool can contextualize documents encompassing not just text but also images, graphs, tables, diagrams, and codes. Furthermore, it is equipped to handle over 100 languages, including Arabic, Japanese, Korean, and French, enabling global enterprises to efficiently access their data.
Cohere also emphasized that Embed 4 has been trained on diverse real-world data, meaning that it can effectively process imperfect documents—such as those with typographical errors, formatting inconsistencies, or varied page orientations—without compromising the accuracy of search results.
Additionally, the AI model possesses specialized knowledge of data from regulated sectors like finance, healthcare, and manufacturing. This allows Embed 4 to be deployed in both VPCs and on-premise environments, ensuring data security for organizations in these industries.
