Cohere's North 2 runs any model, already in production at LG CNS and Bell Cyber

Cohere's North 2 builds on an existing enterprise platform, turning it into a control center for AI agents that handle multi-step workflows on their own and retain context across sessions. The orchestration system coordinates those agents through shared knowledge libraries and reusable skills, while connecting to everyday tools like Slack, SharePoint, and Jira. Rather than locking customers into one model, North 2 is model-agnostic, letting organizations plug in their own models alongside Cohere's own like Command A+. Administrators get granular controls through North Admin, down to token usage and access rights for individual agents, and critical actions require human approval first.

Deployment flexibility appears central to the pitch. Companies can run North 2 on-premises, in the cloud, or fully air-gapped, which matters for government agencies and regulated industries that want to keep AI under their own control. Cohere is targeting that same audience as Aleph Alpha, the Heidelberg-based company Cohere announced it was acquiring in April. Aleph Alpha recently resurfaced with Kolibri, a German-English language model, suggesting the competitive field for sovereign AI in Europe remains active.

Whether North 2 becomes the default control room for enterprise agents likely hinges on adoption beyond the early customers already running North in production. Cohere points to token efficiency on Nvidia's Blackwell and Hopper hardware as one draw, but the bigger question is whether regulated buyers trust the platform's approval and quota controls enough to expand deployments. Companies weighing model-agnostic agent orchestration may see North 2 as a way to avoid vendor lock-in, though the depth of its integrations with tools like SharePoint and Jira will probably decide how deeply it embeds in daily workflows.