Cohere released North 2 on Monday, calling it the biggest upgrade yet to the enterprise agent platform the Canadian company has sold since it went generally available in August 2025. The pitch is blunt: enterprises deploying agents have been forced to trade security against capability against cost, and North 2 claims to end the trade-off.
The release is built around a redesigned agent harness — the orchestration layer that plans multi-step work, calls tools and decides when a human needs to sign off. On top of it sit five named feature groups. Skills package reusable procedures an agent can call on, typically generated by working through a task once and recording which steps produced the best results, then published for colleagues to reuse. Libraries hold the shared knowledge and assets an entire organization draws from. Memory carries context across sessions so an agent does not start cold every time. Applications let users prototype decks, dashboards, documents and lightweight apps from a natural-language prompt. Automations supply prebuilt templates, a drag-and-drop workflow builder and real-time performance monitoring.
The governance layer is where North 2 does its real work. North Admin gives administrators roles and permissions, granular cost controls, rate limits, user quotas and organization-wide caps, with alert thresholds that fire before a limit is hit rather than after. A flow-control view shows which departments are consuming the most tokens, which gives finance teams a concrete target when spend needs cutting. Access control lists attach to libraries and skills individually, so a junior employee prompting for a restricted dataset hits a wall instead of a result.
Deployment is deliberately sovereign. North 2 runs self-hosted, in a customer's virtual private cloud, hybrid, fully on-premises, or completely air-gapped for organizations that permit no outside connection at all. Cohere lists SOC 2 Type 2, ISO 27001 and ISO 42001 certifications, says guardrails screen prompts and responses for personally identifiable information and prompt injection, and points to autonomy policies that restrict agents to actions they are authorized to take. Red-teaming exercises and third-party vulnerability scanning run continuously. The platform is also model-agnostic — customers can bring their own models.
The connector library now covers Slack, SharePoint, OneDrive, Microsoft Outlook, Microsoft Exchange, Jira, Linear, Notion and GitHub, with financial data providers — PitchBook, Crunchbase, Daloopa, FiscalAI, S&P Global and FactSet — listed as planned. Cohere also highlighted an ongoing collaboration with Nvidia, saying its models generate more insights per token and more tokens per second per node when trained and deployed on Blackwell and Hopper hardware.
The reference customers are telling. LG CNS uses North to deliver AI to enterprise customers across South Korea. Bell Cyber runs it inside its own security operations, where Jawad Ahmad, the unit's chief technology and AI officer, said the appeal is applying agentic AI while keeping the control, privacy and data sovereignty its customers demand. Radical Ventures, an early Cohere investor, built an automation that assembles the briefing its partners read before a founder call.
The timing is competitive. OpenAI, Meta and every major cloud vendor have pushed into enterprise agent platforms over the past year, and Cohere's release lands less than three weeks after the company signed a merger agreement with Germany's Aleph Alpha that will keep the Cohere name on the combined company. The differentiator Cohere is betting on is not raw capability — it is the argument that enterprises will not scale agents they cannot govern, and that least-privilege access, token budgets and air-gapped deployment are product features rather than compliance checkboxes. Whether that holds depends on how well administrators configure the controls; an access control list is only as strong as the person who set it up.
Comments (0)
Log in to join the discussion
Log InNo comments yet