Cohere Unveils North 2, Its Enterprise Agent Platform

Canadian AI company Cohere unveiled North 2, its enterprise AI agent platform, on October 5. It redesigned the orchestration system that coordinates multistep agent work and added memory that keeps context across sessions, along with cost controls that let administrators cap token spending. On its official X account, Cohere called the release its "biggest upgrade yet," with more than 15 new features. The biggest change is that the platform now runs not only Cohere's models but also models customers bring from other companies.

Cohere first unveiled North in January 2025 alongside flagship customer RBC, the Canadian bank, and made it generally available that August. Over the past year, Cohere said, it deployed North across finance, healthcare, telecommunications, manufacturing, energy and the public sector, building large-scale production experience that shaped the design of North 2. The company argued that enterprises adopting agents have been forced into a tradeoff — the secure option wasn't smart enough, the smart one wasn't controllable enough, and the governable one became too expensive to run — and that North 2 ends that tradeoff.

The new orchestration system handles multistep tasks on its own and brings in a human only for steps that need sign-off. Agents and automations built from simple prompts can be shared across the company, so other teams don't have to rebuild the same workflow from scratch. Logic that agents call on repeatedly is packaged as skills, and documents and assets every agent should reach go into shared libraries. Users can prototype slides, dashboards, documents and lightweight apps in natural language inside the chat window, and recurring work gets prebuilt templates and a drag-and-drop visual workflow builder.

Cohere's 1-minute-35-second launch video, which Metal reviewed, shows that flow. A user asks for a corporate credit risk dashboard to take in, check and analyze customer credit data; North produces a dashboard app linked to the data and reorganizes it with an executive-overview skill. The video then schedules a commercial credit risk assessment automation for Friday and shows an admin screen turning on PII detection for prompts and responses as well as prompt injection detection.

Cohere also widened its reach. New connectors cover Slack, SharePoint, OneDrive, Outlook, Exchange, Jira, Linear, Notion and GitHub, and connections to financial data providers such as PitchBook, Crunchbase, S&P Global and FactSet are planned. Security is applied per agent. In the North Admin console, guardrails are set for each agent, autonomy policies keep each agent to the actions it is authorized to take, and critical decisions call for human oversight. Deployment options range from self-hosted servers and a customer's virtual private cloud to fully air-gapped environments with no outside connection, and the platform holds SOC 2 Type 2, ISO 27001 and ISO 42001 certifications.

Cost control is the other pillar of the release. North Admin tracks token use down to the individual user and agent, and administrators can define consumption tiers for users and groups based on request and token rates and cap usage across the company. Alerts can warn teams before a limit is reached. According to reports, Paul Teyssier, Cohere's vice president of product for AI, said cost control is "probably the number one rising concern for AI leaders," adding that Cohere has not yet built its own model router into the platform but is seeing demand for it.

Opening up to other models is a decision that could reduce use of Cohere's own models. According to reports, North 2 customers can plug in the most advanced models from Anthropic, OpenAI and Google, or cheaper open-source models from Chinese firms such as Alibaba, DeepSeek and Z.ai. "A whole bunch of players [are] trying to take the customers deeper and deeper into their ecosystem, with no way out," Teyssier said. "We are intentionally taking the other approach." He added that customers are "going to prefer our models based on hard facts, not based on the fact that it's locked together." Cohere said its work with NVIDIA means its models produce more tokens per second per node on Blackwell and Hopper GPUs, and Kari Briski, NVIDIA's vice president of generative AI, said this helps North 2 "deliver more tokens for less."

Customer examples came with the launch. "North gives enterprises a secure, practical foundation for turning agentic AI into real business value," said Yohan Jin, head of the AI Center at LG CNS. LG CNS uses North internally and also brings it to customers across Korea. Jawed Ahmad, chief technology and AI officer at Bell Cyber, said North lets the unit bring agentic AI into its security operations with the control, privacy and sovereignty requirements its customers demand built in from the start. Radical Ventures, an early Cohere investor, built an automation on North that prepares the briefing its team reads before a call with a founder.

Metal has reported on Cohere's launch of the document parsing model Parse 5, and this announcement came less than three weeks after Cohere signed a merger agreement with German AI company Aleph Alpha. The combined company will keep the Cohere name. According to reports, other North customers include telecom company Bell, Saudi telecom STC Group and the Aston Martin F1 team.

From an AI engineer's perspective, the weight of North 2 sits not in the model but in the harness. Just as coding agents changed how developers work, Cohere wants to bring the same shift to non-developer roles such as finance and marketing by handing them prebuilt skills and agent libraries. "The product itself needs to do a really good job at helping people go up the AI adoption curve," Teyssier said. By leaving models swappable while keeping memory, permissions and the token ledger on its own platform, Cohere has staked its next phase of growth on owning the agent operations layer for heavily regulated industries.