Cohere North 2 Brings Token Spending Caps and Memory to Enterprise AI Agent Platform

Cohere North 2 Addresses Enterprise AI Cost and Control Challenges

Cohere Inc. unveiled North 2, a major overhaul of its enterprise AI agent platform designed to eliminate the tradeoff between security, capability, and cost control that organizations have accepted when deploying AI agents3

3

. The Canadian model developer introduced the update on Monday, building on a year of large production deployments across finance and public sector industries since North became generally available in August 20253

3

. The platform competes directly with Microsoft Copilot and Google Vertex AI in the enterprise AI market2

2

.

Rebuilt Orchestration System and Shared Agent Memory Transform Workflows

The agent harness features a redesigned orchestration system that handles complex multi-step processes autonomously while keeping humans in the loop for designated decisions2

2

. North 2 introduces shared agent memory that maintains context across sessions, preventing agents from starting cold each time2

2

. Paul Teyssier, VP of Product AI at Cohere Inc., explained that memory lets context persist longer across sub-sessions and agents, though specific duration figures were not disclosed2

2

.

The platform now supports reusable skills and automations that teams can share organization-wide, eliminating redundant workflow development1

1

. Skills function as standard operating procedures documenting how the agent harness should complete tasks, often generated by working with the agent and documenting which steps achieved optimal results1

1

. Libraries hold shared knowledge and assets accessible across the organization, ensuring agents work from consistent datasets2

2

.

Token Spending Caps and Flow Control Tackle Budget Overruns

North 2 implements comprehensive token spending caps through North Admin, the platform's management console3

3

. Administrators can track token usage down to individual users and agents, define consumption tiers based on request and token rates, and set organization-wide caps2

2

. Flow control functionality provides visibility into departmental token consumption, while alert thresholds warn teams before limits apply1

1

. Usage data can be broken down by model, user, and department, then exported into customers' own analytics systems2

2

.

These controls address a critical gap identified in VB Pulse's agentic orchestration tracker from July, which found that 21% of enterprises exercise only reactive monitoring of agent spend with no real-time programmatic way to prevent budget-breaking bills2

2

. Competing platforms from Microsoft, OpenAI, and Anthropic have added similar spending controls, though these typically govern usage only within each provider's own platform2

2

.

Access Control Lists and Security Guardrails Strengthen Enterprise Protections

North 2 implements granular access control lists that administrators can assign to libraries and skills, preventing unauthorized access to proprietary data1

1

. Security guardrails scan prompts and responses for PII screening and detect prompt injection protection, with enforcement at the individual agent level2

2

. Autonomy policies limit agents to authorized actions and route critical decisions to humans3

3

.

The platform maintains SOC 2 Type 2, ISO 27001, and ISO 42001 certifications, with continual security testing including red-teaming and third-party vulnerability scanning2

2

. Jacob Rideout, CTO of HiddenLayer, noted his team only activated agent workflows pulling untrusted data from marketing, engineering, and security operations once they could see and control agent actions2

2

.

Sovereign AI and On-Premise Deployment Options Reduce Vendor Lock-In

Sovereign AI remains central to Cohere's positioning, with North 2 supporting deployment in cloud environments, on-premise installations, or fully air-gapped configurations1

1

. The model-agnostic architecture allows enterprises to use Cohere's models or their preferred alternatives, with administrators assigning specific models to teams3

3

. Teyssier emphasized that North 2's harness, search, and embedding components are interchangeable, reducing vendor lock-in2

2

.

Cohere's partnership with Nvidia aims to reduce operational costs, with the company stating its models produce more tokens per second per node on Nvidia's Blackwell and Hopper GPUs3

3

. Kari Briski, VP of generative AI at Nvidia, confirmed this helps North 2 deliver more tokens for less3

3

.

Early Adopters Report Productivity Gains Across Industries

Bell Canada's Bell Cyber unit deployed North within its security operations, with CTO Jawed Ahmad stating the platform meets the control and sovereignty requirements Bell Cyber's customers demand3

3

. South Korean IT services company LG CNS Co. Ltd. runs North internally and offers it to customers nationwide3

3

. Radical Ventures, an early Cohere investor, built an automation that prepares briefings before founder calls, eliminating hours of manual context gathering according to partner Parasvil Patel3

3

.

The release arrives less than three weeks after Cohere signed a merger agreement with German AI developer Aleph Alpha GmbH, with the combined entity retaining the Cohere name3

3

. In VB Pulse's July agentic orchestration survey of 61 respondents, Microsoft led at 41%, followed by Anthropic at 28%, LangChain/LangGraph at 10%, and OpenAI at 7%2

2

. Watch for Cohere to leverage its model-agnostic approach and deployment flexibility as differentiators against platform-locked competitors in the intensifying enterprise AI agent market.