Cohere launched North 2, the new version of its enterprise platform for AI agents, in an update focused on running multistep tasks, managing costs, maintaining security controls, and connecting agents to the systems used by companies. The company says the release represents the platform’s biggest update since North was made generally available in August 2025.
Agent Orchestration and Context Retention
North 2 relies on a new agent orchestration system capable of managing complex, multistage processes. Users can create agents and automations through natural-language commands, while keeping certain points under human supervision when a process requires approval or a sensitive decision. Agents can also be shared across the enterprise, reducing the need for different teams to rebuild the same workflows.
The platform adds the Memory feature to retain context between sessions, while Skills enables the reuse of recurring capabilities and business logic. Libraries provides shared repositories for company-wide information and files. Within the chat interface, users can create presentations, documents, dashboards, and prototypes for simple applications using natural-language commands.
Integrations and Monitorable Automation
North 2 includes ready-made templates for recurring processes, along with a visual tool for building workflows using drag and drop. It also enables real-time monitoring of automation performance. Available integrations include Slack, SharePoint, OneDrive, Microsoft Outlook, Microsoft Exchange, Jira, Linear, Notion, and GitHub.
Cohere said it is also working on integrations with financial data providers PitchBook, Crunchbase, Daloopa, FiscalAI, S&P Global, and FactSet, without specifying when they will become available.
More Precise Control Over Token Consumption
North 2 focuses on an aspect that often becomes more complex as the number of agents within an enterprise increases: controlling usage costs. The North Admin dashboard enables token consumption to be tracked at the user and agent levels, with different usage levels set for users and teams according to the number of requests or the volume of consumption.
Organizations can also set overall token-consumption limits and send advance alerts when they approach them. In practice, these tools give administrative teams a way to monitor automation costs before they turn into unexpected spending, but they do not eliminate the need to establish internal policies for choosing models and assigning appropriate tasks to each agent.
Deployment, Security, and Model Compatibility
North 2 is model-independent; companies can use Cohere models or integrate other models of their choice. On the security level, the platform provides permissions policies and rules for each agent, including autonomy policies that restrict the operations an agent is allowed to perform. Human approval can also be required for sensitive decisions or actions.
The platform supports operation within a company’s private infrastructure, in a private cloud, or within air-gapped systems that are not connected to the internet. This makes the offering particularly targeted at sectors such as financial services, government, healthcare, and telecommunications, where sensitive data and regulations impose strict requirements on where systems operate and how they are accessed.
What Changes in Practice?
North 2 is not limited to adding a conversational agent; it attempts to bring agent creation, distribution, performance monitoring, and cost control together within a single enterprise layer. According to the article, Cohere is drawing on its experience with large-scale deployments in the finance, healthcare, telecommunications, manufacturing, energy, and government sectors. It also said that its collaboration with Nvidia improved token-generation speed for its models on Blackwell and Hopper units, which could help reduce usage costs.
However, the available information does not include figures for North 2 pricing, actual consumption limits, or performance-measurement details. Therefore, the value of the release for organizations remains tied to their ability to integrate their existing systems, control agent permissions, and assess the deployment’s impact on cost and compliance in their own environments.