Cohere announces North 2 agent harness for enterprise AI security

Canadian model developer Cohere announced a revamped version of its North agent harness, aiming to provide enterprises with a safer and more consistent framework for deploying autonomous business processes, The Register reported. While competitors such as OpenAI and Anthropic focus heavily on rogue agent fears, Cohere designed North 2 to integrate tightly with existing knowledge bases and SaaS apps while enforcing stricter access controls and token spending limits.

North 2 Features Shared Skills and Memory

North 2 introduces several concepts that mirror tools like Claude Code and OpenClaw, including skills, libraries, automations, and memory. Skills act as standard operating procedures that document how an agent harness should complete tasks, often generated by working alongside the agent to record optimal steps. Cohere stated that these skills are now easier to share across an organization to prevent users from reinventing workflows. Automations supply generic templates customizable via a drag-and-drop graphical interface resembling LangFlow, while libraries establish shared knowledge assets to ensure agents work from identical datasets. The platform also adds memory to let agents maintain context across multiple sessions.

Enterprise Security and Sovereign Deployment Options

Sovereign AI remains a core focus for Cohere. Enterprises can deploy North 2 on-premises in air-gapped environments or run it via the cloud using their preferred models. Alongside deployment flexibility, North 2 implements guardrails designed to manage resource consumption. Flow control functionality offers visibility into token usage across different departments, while access control lists let administrators restrict access to specific libraries and skills so that internal data remains protected.

Crowded Enterprise Agentic Market

North 2 arrives amid a wave of enterprise-focused agentic AI platforms launched over the past year. Major and minor cloud providers have introduced tailored agent harnesses as competition intensifies, particularly with OpenAI, Meta, and others targeting enterprise customers.