Cohere North: Secure AI Agent Platform for Enterprises

Addressing Data Security Concerns with AI Agent Tools

The potential of AI agent tools to streamline workflows is significant. However, widespread adoption is currently hampered by substantial concerns regarding data security within many organizations.

Enterprises safeguarding trade secrets, those operating in heavily regulated sectors, and governmental bodies are proceeding with caution. The risk of inadvertent data compromise, or the potential for sensitive information to be utilized in the training of large language models, is a major deterrent.

Cohere's North: A Private AI Agent Platform

Cohere, a Canadian AI company, is introducing North, a new AI agent platform designed to mitigate these security risks. North is engineered to facilitate private deployment, ensuring that both organizational and customer data remains protected by existing security infrastructure.

According to Nick Frosst, co-founder of Cohere, the effectiveness of Large Language Models (LLMs) is directly tied to the data they can access. To maximize utility, LLMs must be deployed within the customer’s own environment.

Deployment Flexibility and Infrastructure Requirements

Unlike solutions reliant on public cloud platforms such as Azure or AWS, Cohere asserts that North can be installed directly on an organization’s private infrastructure. This ensures that customer data is never transmitted to or processed by Cohere.

North’s deployment options are versatile, encompassing on-premise infrastructure, hybrid cloud setups, Virtual Private Clouds (VPCs), and even completely air-gapped environments, as Frosst explained.

The platform is designed for accessibility, capable of running on minimal hardware. “We can deploy literally on a GPU in a closet that they might have somewhere,” Frosst stated, noting that as few as two GPUs are sufficient to operate North.

Robust Security Features and Compliance

Beyond private deployment, Cohere emphasizes that North incorporates a comprehensive suite of security measures. These include:

• Granular access control
• Agent autonomy policies
• Continuous red-teaming exercises
• Third-party security assessments

Furthermore, North is designed to meet stringent international compliance standards, including GDPR, SOC-2, and ISO 27001, demonstrating a commitment to data protection and regulatory adherence.

Expanding Beyond Private Deployments

Cohere, having secured $970 million in funding, with a recent valuation of $5.5 billion, has announced initial trials of North with several key clients. These include RBC, Dell, LG, Ensemble Health Partners, and Palantir, as previously covered by TechCrunch.

North presents a suite of features common to many AI agent platforms. Specifically, it offers chat and search capabilities, enabling users to address customer support questions, condense meeting records, generate marketing materials, and retrieve data from both internal systems and the wider internet.

All responses generated by the platform are accompanied by source citations and detailed “reasoning” pathways. This allows employees to review and validate the information provided.

The chat and search functionalities leverage Cohere’s existing technologies. These include the Command family of generative AI models and the Compass multimodal search tech stack.

Frosst explained that North utilizes a specialized version of the Command model, specifically optimized for enterprise-level reasoning tasks.

“The platform’s capabilities extend beyond simple question answering to encompass task completion,” Frosst stated. “North facilitates asset creation, including tables, documents, and slideshows, and can perform extensive market research.”

Notably, Cohere completed the acquisition of Ottogrid in May. Ottogrid is a Vancouver-based company specializing in enterprise tools for automating advanced market research.

Similar to other AI agent platforms, North integrates with popular workplace applications. These include Gmail, Slack, Salesforce, Outlook, and Linear.

Furthermore, it supports integration with any servers utilizing the Model Context Protocol (MCP). This allows access to specialized industry applications or proprietary internal tools.

“Users can gradually increase their reliance on the model,” Frosst noted. “This transition occurs naturally, moving from using North as a supportive tool to employing it for automated processes.”

Note: A prior iteration of this article contained an inaccuracy regarding Frosst’s official title. We apologize for this error.