Cohere has released North 2, a major upgrade to the company’s agentic AI platform. The platform features a new orchestration system for agents, a memory system, and comprehensive management tools to limit token consumption. Unlike the largest AI model builders, Cohere also supports deployments in air-gapped environments and on-premises.
North was launched in August of last year as a platform that keeps corporate data within the organization’s own environment. Since then, Cohere has rolled it out in sectors including finance, healthcare, telecommunications, energy, and the public sector. North 2 is built on that practical experience.
Agents that remember tasks
The foundation is a new agent harness designed to independently handle complex, multi-step processes. Users build agents and automations using prompts and share them within the organization. Skills, which, much like elsewhere, are essentially reusable capabilities, play a crucial role here. Other key features include Libraries for shared knowledge and Memory, which allows agents to retain their context between sessions. Additionally, North generates dashboards, documents, and basic apps directly from the chat function.
New connectors further integrate the platform with Slack, SharePoint, OneDrive, Outlook, Exchange, Jira, Linear, Notion, and GitHub. Financial data sources such as PitchBook, Crunchbase, S&P Global, and FactSet will follow later.
Control over tokens
Through North Admin, administrators can set roles, rate limits, per-user quotas, and organization-wide caps. Cohere promises that alerts will notify users before a limit is reached. The platform is model-agnostic, so “Bring Your Own Model” gives users flexibility. And according to both companies, those running Cohere models on Nvidia Blackwell or Hopper will achieve more tokens per second per node.
As mentioned, Cohere’s deployment options are very broad. These include self-hosted, VPC, hybrid, on-premises, or fully air-gapped environments. Guardrails operate down to the level of individual agents and can therefore be customized based on the use case.
Read also: Cohere and Aleph Alpha merge: an AI company with two headquarters