Cohere launches North 2, enhancing AI agents with memory and controls
Summary
Cohere has launched North 2, an upgraded version of its enterprise agent platform that addresses common issues such as budget management and contextual accuracy in AI agents. With features like user quotas, rate limits, and a memory system that maintains context across sessions, North 2 allows organizations to have tighter control over their AI spending and improves the reliability of agent outputs, which many enterprises have reported as problematic due to inconsistent business context. This release illustrates a shift in the market as companies increasingly prioritize security, manageability, and flexibility in their AI deployments, with a growing demand for real-time tools to monitor agent spending actively.
Analysis
Cohere: Cohere is an enterprise AI company specializing in language models and agent orchestration platforms designed for secure, customizable deployments. North 2 represents its updated enterprise agent platform, incorporating memory for cross-session context, user quotas, rate limits, organization-wide caps, and a redesigned harness for multi-step processes. The platform supports cloud, on-premises, and air-gapped environments while integrating with existing identity systems and emphasizing model portability to minimize lock-in. Paul Teyssier: Paul Teyssier is Vice President of Product AI at Cohere. He discussed North 2's memory feature enabling persistent context across sub-sessions and agents, as well as enhanced granular controls for managing automations and reporting at scale. Teyssier highlighted the focus on enterprise-level oversight rather than mere repeatability of processes. Context and Accuracy: Many organizations report issues with confidently incorrect agent outputs stemming from inconsistent or missing business context. Agent Spend Management: Enterprises are seeking real-time programmatic tools to monitor and cap AI agent spending to prevent unexpected costs. Deployment and Control: Vendors are adding spend controls and hybrid options as enterprises prioritize security, permissioning, and reduced vendor lock-in in AI agent platforms.
Categories
aiai_agentstechmachine_learningvirtuals