Cohere Launches Secure Enterprise AI Platform North

AI agent tools hold the potential to reduce repetitive tasks in daily workflows, yet many organizations remain cautious about adoption. A primary concern is data security. Large enterprises with proprietary secrets, firms in heavily regulated sectors, and government bodies have significant reservations. They worry that their own data—or their customers'—could be exposed or used to train public AI models without proper safeguards.

Canadian AI company Cohere is addressing these concerns with its new AI agent platform, North. It's designed for private deployment, allowing enterprises and governments to keep their sensitive data securely behind their own firewalls.

"Large language models are only as effective as the data they can access," explained Cohere co-founder Nick Frosst during a North demonstration. "To maximize their usefulness, they need access to relevant information, which often requires deployment directly within the customer's own IT environment."

Rather than operating on public cloud platforms like Azure or AWS, Cohere can install North directly on an organization's private infrastructure. This approach ensures Cohere never interacts with or views customer data. According to Frosst, North can run on on-premise servers, hybrid clouds, virtual private clouds, or even isolated air-gapped systems.

"We can deploy it on a GPU in a closet somewhere on their premises," he noted, adding that North is engineered to operate efficiently on a minimal setup of just two GPUs.

Cohere states that North incorporates robust security measures, including granular access controls, policies governing agent autonomy, continuous security testing, and third-party audits. The platform also complies with international standards such as GDPR, SOC-2, and ISO 27001.

More than private deployments

Cohere, which has raised $970 million to date and was last valued at $5.5 billion, has already conducted pilot programs for North with clients including RBC, Dell, LG, Ensemble Health Partners, and, as previously reported by TechCrunch, Palantir.

Leading Tech and VC Figures Join the Disrupt 2025 Agenda

Hear from top voices at Netflix, ElevenLabs, Wayve, and Sequoia Capital, who will be joining the Disrupt 2025 agenda. They'll share critical insights to drive startup growth and sharpen your competitive edge. Don't miss the 20th anniversary of TechCrunch Disrupt—secure your ticket today and save up to $675 before the price increases on August 7.

Leading Tech and VC Figures Join the Disrupt 2025 Agenda

Hear from top voices at Netflix, ElevenLabs, Wayve, and Sequoia Capital, who will be joining the Disrupt 2025 agenda. They'll share critical insights to drive startup growth and sharpen your competitive edge. Don't miss the 20th anniversary of TechCrunch Disrupt—secure your ticket today and save up to $675 before the price increases.

North offers a range of standard AI agent capabilities. Its core features include chat and search functions that allow users to get answers to support questions, summarize meeting notes, draft marketing content, and pull information from internal databases and the web. Frosst emphasized that all responses come with source citations and transparent reasoning steps, enabling employees to review and verify the AI's output.

These functions are built on Cohere's existing technology, such as its Command generative AI models and its Compass multimodal search platform. Frosst mentioned that North utilizes a specialized version of the Command model, fine-tuned for complex enterprise reasoning tasks.

"It moves beyond simple question-and-answer to actually performing work. North can create various assets—tables, documents, slide presentations—and conduct in-depth market research," Frosst added.

It's notable that Cohere recently acquired Ottogrid, a Vancouver-based platform specializing in automated, high-level market research tools for enterprises.

Similar to other agent platforms, North can connect to common workplace applications like Gmail, Slack, Salesforce, Outlook, and Linear. It also integrates with Model Context Protocol servers to connect with specialized or custom internal software.

"As users build trust through interacting with the model, there's a natural progression from using it as a productivity aid to leveraging it for full task automation," Frosst observed.

Correction: A previous version of this article misstated Frosst's title. We regret the error.