Cohere Inc. announced North 2 on October 5, 2026. The version replaces the earlier North system that has been generally available since August 2025. North 2 lets large firms run AI agents that can handle multi‑step tasks, pull in a human when needed, and keep memory between sessions. Administrators now see token use for each user and can set spending limits or alerts directly in the North Admin console.
What’s new in North 2?
Agents can be built from a simple prompt and then shared across the organization, avoiding duplicate work. A new “skills” feature lets teams package reusable logic that agents call repeatedly. Shared libraries store documents and other assets that any agent can access.
From a plain language request, an agent can draft a slide deck, generate a dashboard, or spin up a lightweight app inside the chat window. Pre‑built templates and a drag‑and‑drop workflow builder help set up recurring processes, and performance metrics appear in real time. The connector library now includes Slack, GitHub, SharePoint and OneDrive, with plans for financial data sources such as PitchBook, S&P Global and FactSet.
Security and autonomy controls sit in North Admin. The system screens for personally identifiable information, blocks prompt‑injection attempts, and limits each agent to the actions it is authorized to take. North can run on‑premises, in a private cloud, or in fully air‑gapped environments.
Why the token caps matter
Cohere says the new token‑spending caps give enterprises a way to control AI costs without sacrificing capability or security. The caps are defined per user or group, and alerts warn teams before they hit their limits. Because the platform is model‑agnostic, companies can also bring their own models and still benefit from the same cost‑control tools.
Why it matters
For IT managers, the token caps give a concrete way to prevent runaway AI spending while still allowing teams to automate complex workflows. For end users, shared agents and built‑in templates mean they can get AI‑generated reports or apps faster without building the same logic from scratch. What remains unclear is how much the caps will affect performance for heavy‑use cases, since pricing details were not disclosed.
