Cohere's North AI Agent Platform Secures Enterprise Data On-Premise
The widespread promise of AI agent tools, designed to streamline daily workflows and alleviate drudgery, has been met with significant hesitation from organizations grappling with a critical concern: data security. Large enterprises safeguarding valuable trade secrets, companies operating in highly regulated sectors, and government agencies have all approached AI adoption with extreme caution, fearing that their proprietary information—or worse, their customers’ sensitive data—could be inadvertently compromised or exploited to train foundational AI models.
Canadian AI firm Cohere is directly addressing these anxieties with its new AI agent platform, North. The platform is designed to enable private deployment, ensuring that enterprises and governments can keep their and their clients’ data securely behind their own firewalls. Nick Frosst, co-founder and CEO of Cohere, emphasized this during a demonstration of North, stating, “LLMs are only as good as the data they have access to. If we want LLMs to be as useful as possible, they have to access that useful data, and that means they need to be deployed in [the customer’s] environment.”
Unlike solutions that rely solely on public enterprise cloud platforms like Azure or AWS, Cohere asserts that North can be installed directly onto an organization’s private infrastructure. This approach means Cohere itself never accesses or interacts with a customer’s data. North offers versatile deployment options, capable of running on an organization’s on-premise infrastructure, various hybrid cloud setups, virtual private clouds (VPCs), or even highly secure air-gapped environments. Frosst highlighted North’s minimal hardware requirements, noting its ability to function effectively on as few as two GPUs, humorously adding, “We can deploy literally on a GPU in a closet that they might have somewhere.”
Beyond its flexible deployment, North integrates robust security protocols. These include granular access control, which allows precise management of who can access what data; agent autonomy policies, dictating how AI agents operate; continuous red-teaming, a proactive security testing method; and third-party security evaluations. The platform also boasts compliance with major international standards such as GDPR, SOC-2, and ISO 27001, further cementing its commitment to data integrity and privacy.
Cohere, which has successfully raised $970 million and recently achieved a valuation of $5.5 billion, has already piloted North with several prominent clients, including RBC, Dell, LG, Ensemble Health Partners, and Palantir. These early adoptions underscore the platform’s practical applicability in diverse enterprise settings.
At its core, North offers key AI agent functionalities such as sophisticated chat and search capabilities. These features enable users to efficiently manage customer support inquiries, accurately summarize meeting transcripts, generate marketing copy, and seamlessly access information from both internal company resources and the broader web. A crucial transparency feature is North’s inclusion of citations and “reasoning” chains for all responses, allowing employees to audit and verify the AI’s output. The chat and search functions are powered by Cohere’s existing technological suite, including Command, its family of generative AI models, and Compass, its multimodal search technology. Specifically, North utilizes a variant of the Command model optimized for complex enterprise reasoning.
The platform extends beyond simple question-and-answer interactions, venturing into active work execution. North facilitates significant asset creation, capable of generating tables, documents, slideshows, and conducting comprehensive market research. This enhanced capability was bolstered by Cohere’s acquisition of Ottogrid in May, a Vancouver-based platform specializing in automating high-level market research for enterprises. Furthermore, North is designed for seamless integration with a wide array of existing workplace tools, including popular applications like Gmail, Slack, Salesforce, Outlook, and Linear. It can also connect with Model Context Protocol (MCP) servers, enabling access to industry-specific or in-house applications. Frosst envisions a natural progression for users: “As you build confidence by chatting to the model, there’s like a smooth transition that happens between using this as an augmentation to using it as an automation.” This vision highlights North’s potential to evolve from a supportive tool into a truly transformative force for enterprise operations.