Cohere released North 2, a major overhaul of its platform for running AI agents inside large organizations, adding a rebuilt orchestration system, persistent agent memory and controls that let administrators cap token spending, SiliconANGLE reported Monday. The Toronto-based company said a year of large deployments since North became generally available in August 2025 shaped the update.

What Happened

The new orchestration layer breaks multistep tasks into stages, carries them out and pulls in a person when a step requires approval. Agents built from simple prompts can now be shared across a company, and they retain memory between sessions instead of starting fresh each time.

Cohere also added "skills," which package reusable agent logic, and shared libraries for documents and other assets. Users can ask North in plain language to produce slide decks, dashboards and lightweight applications. Prebuilt templates and a drag-and-drop workflow builder report on agent performance as tasks run.

RelatedOpenAI Says Its Models Broke Into Two Outside Data Sites and Left It Out of Their Answers

On the cost side, the North Admin console now tracks token consumption by individual user and by agent. Administrators can define consumption tiers, set a ceiling on company-wide usage and receive alerts before limits are reached.

The platform remains model-agnostic, so customers can assign outside models to particular teams. New connectors cover Slack, GitHub, Microsoft SharePoint and OneDrive, and Cohere plans to add feeds from financial data providers PitchBook, S&P Global and FactSet.

Guardrails screen prompts and responses for personal information and prompt injection attempts and enforce policies on how much autonomy agents have. North can still be deployed on premises, inside a customer's virtual private cloud or in fully air-gapped environments.

Cohere also said it worked with Nvidia to improve how many tokens per second North generates on the chipmaker's Blackwell and Hopper GPUs.

Why It Matters

Spending controls address one of the less visible problems of agent rollouts. Agents that plan, call tools and retry consume far more tokens than a single chatbot exchange, and costs can climb quickly once hundreds of employees build their own. Per-user and per-agent metering gives finance and IT teams a way to see where usage comes from and to stop overruns before the bill arrives.

The deployment options matter for Cohere's target market. The company has focused on banks, telecom operators and public-sector bodies that cannot send sensitive data to a shared public cloud, and the air-gapped option is aimed at buyers with strict sovereignty or classification rules.

Customers and Merger

SiliconANGLE named several users. Bell Canada's Bell Cyber unit uses North in security operations, and its chief technology and AI officer, Jawed Ahmad, said the platform's control and sovereignty features were built in from the start. South Korea's LG CNS runs North internally and resells it to Korean customers. Venture firm Radical Ventures built an automation that prepares briefings before calls with founders, cutting work that used to take hours.

North 2 arrives less than three weeks after Cohere agreed on Sept. 16 to merge with Germany's Aleph Alpha, with the combined company keeping the Cohere name. Aleph Alpha has focused on sovereign AI for European governments and industry, an overlap with North's private-deployment pitch.

What to Watch

Key follow-ups are the timing of the PitchBook, S&P Global and FactSet connectors, which would make North more useful to the financial firms Cohere courts, and how the platform is folded together with Aleph Alpha's products once the merger closes. Customer reports on whether the token caps actually rein in agent costs will show whether the controls work in practice.