AI Briefing
KOSign in

Cohere launches North 2, a model-agnostic agentic AI platform for enterprise sovereignty

·2026.10.05 22:04

Key point

North 2 introduces a redesigned agent harness, granular cost governance via North Admin, and support for on-premises and air-gapped deployments.

Details

Cohere has released North 2, an upgrade to its enterprise agentic AI platform designed to eliminate the trade-offs between security, intelligence, and cost. The platform unifies orchestration, governance, and model flexibility into a single system that can run wherever an organization's data resides, including on-premises and air-gapped environments.

Enhanced Agentic Capabilities

The core of North 2 is a redesigned agent harness that enables sophisticated, multi-step automations. Key features include:

  • Skills and Libraries: Reusable capabilities and shared knowledge assets to prevent redundant workflow creation.
  • Memory: Agents retain context across sessions rather than starting fresh.
  • Applications: Users can prototype decks, dashboards, and lightweight apps directly from natural language prompts.
  • Connectors: New integrations with Slack, SharePoint, OneDrive, Microsoft Outlook, Exchange, Jira, Linear, Notion, and GitHub, with planned support for financial data providers like PitchBook and S&P Global.

Enterprise Governance and Security

North 2 introduces North Admin, a control layer for granular cost management and security. Admins can set rate limits, user quotas, and org-wide caps to ensure predictable AI operations. The platform supports SOC 2 Type 2, ISO 27001, and ISO 42001 certifications, and offers deployment flexibility across self-hosted, VPC, hybrid, and on-premises environments. Guardrails and autonomy policies allow teams to restrict agent actions and require human oversight for critical decisions.

Model Agnosticism and Efficiency

The platform is model-agnostic, allowing enterprises to use Cohere’s models or bring their own. Cohere highlights a collaboration with NVIDIA, noting that using Cohere models trained and deployed on NVIDIA Blackwell and Hopper hardware increases tokens per second per node, thereby reducing spend without sacrificing intelligence. This efficiency aims to improve AI ROI for regulated industries such as finance, healthcare, and the public sector.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.