Cohere North 2 was released on October 5, 2026: enterprise AI company Cohere announced North 2 that day, describing it officially as the biggest upgrade to the North platform to date, bringing more than 15 new features at once and combining security, intelligence, cost governance, and full-stack control in a single agent platform. The previous generation of North became generally available in August 2025, with a focus on deploying agents inside an enterprise's own environment; what North 2 adds this time is not model intelligence, but the ledger that corporate finance and IT departments care about most.
The most practical addition: a token cap for every employee
North 2 adds the ability to view token usage by user and by team in the admin console, and administrators can set spending caps and alerts when usage approaches those caps directly. Once agents are running inside an enterprise, runaway usage bills are a common concern: an automation workflow that gets triggered repeatedly can burn through more in a single night than an entire team's monthly budget. Caps and alerts amount to a master switch for agents, which is also why industry outlets such as SiliconANGLE put "token spending caps" in their headlines.
Another detail is that the platform is not tied to Cohere's own models: when an enterprise connects third-party models, the same usage tracking and cost controls still apply. If cost governance only worked for its own models, large enterprises would never buy it. This point determines that North 2's ledger features are a platform-level capability, not an add-on for promoting its own models.
Agents shift from personal hand-built tools to assets shared across the whole company
In the previous generation of North, agents were mostly used by whoever built them. North 2 lets employees generate an agent from a single natural-language description and share it across the entire organization, so others do not have to reinvent the wheel; the newly added Skills package reusable processing logic into skills that different agents can call repeatedly, while Libraries store documents and materials shared across the organization. Agents have also gained cross-session memory, so they do not have to understand context from scratch every time, and when executing multi-step tasks they can bring a person back in to confirm at key points.
Output is also more direct: presentations, dashboards, documents, and even lightweight applications can be generated right in the chat window; prebuilt templates and a drag-and-drop workflow builder turn repetitive work into automated processes, with real-time metrics showing how they run. The connector library has added Slack, GitHub, SharePoint, and OneDrive, and financial data sources such as PitchBook, S&P Global, and FactSet are still planned for integration.
Guardrails and deployment tighten together, with the selling point still "data never leaves"
Governance features are concentrated on the admin side: the system screens for personally identifiable information, blocks prompt injection attempts, and limits each agent's actions to its authorized scope. The risks of enterprise agents are no longer hypothetical — Databricks Genie was just disclosed to be vulnerable to a malicious Skill bypassing four types of controls to carry out phishing and data exfiltration — and North 2 puts skills, permissions, and auditing under governance at the same layer, a prescription aimed directly at this kind of incident.
Deployment continues Cohere's long-standing selling point: it can run on-premises, in a private cloud, and even in fully isolated environments with no external network access, combined with a zero-trust framework, fine-grained access controls, and auditable run records, targeting industries where data cannot leave the premises, such as finance, healthcare, and government.
The cold water to pour here is price: Cohere has not disclosed specific pricing for North 2, and officials have given no data on whether strictly set caps will slow down heavy tasks. For enterprise IT teams currently evaluating options, North 2's differentiation is not the model, but "whether you dare let agents run freely across the whole company" — while others compete on how much agents can do, Cohere is competing this time on whether they can be controlled before something goes wrong and whether the bill can be explained after it arrives.
