Cohere's North 2 rebuilds its enterprise agents around control

Cohere has rebuilt its enterprise AI platform from the harness up. North 2, published on 5 October, replaces the orchestration layer that drives agents and automations, adds memory that survives between sessions, and hands administrators the cost controls that decide whether an agentic rollout is affordable.

A harness for agents, not just a chat window

At the centre of North 2 is a new agent harness, Cohere's term for the software that plans and runs multi-step tasks. Agents and automations can now be built from prompts and then shared across an organisation as reusable components, so the same workflow is not rebuilt team by team. Four ideas carry that: Skills, reusable capabilities an agent can call on; Libraries, shared knowledge and assets; Memory, so a session does not start cold; and Applications, which let users prototype decks, dashboards, documents and lightweight apps in natural language. Automations ship with templates, a drag-and-drop visual builder and monitoring, and new connectors tie North to Slack, SharePoint, OneDrive, Outlook, Exchange, Jira, Linear, Notion and GitHub, with financial data providers including PitchBook, Crunchbase and S&P Global planned.

Sovereignty is the pitch

North 2 is model-agnostic, so teams can use Cohere's own models or bring their own. It can be deployed self-hosted, in a VPC, hybrid, on-premises or fully air-gapped, and Cohere says the platform holds SOC 2 Type 2, ISO 27001 and ISO 42001 certifications. North Admin adds granular cost controls, rate limits, user quotas and organisation-wide caps, along with autonomy policies that restrict what an agent is authorised to do and push critical decisions to a human. Cohere points to LG CNS in South Korea and Bell Cyber as organisations already running North in production.

Why enterprises get stuck

The trade-off Cohere is describing will be familiar to anyone who has tried to put agents inside a regulated company. A platform secure enough to clear a security review is rarely the smartest one; the models that reason well are hard to govern; and the governable ones get expensive at scale. North 2 is Cohere's claim that all four properties can sit in the same box. There is no public pricing, and Cohere directs interested buyers to its sales team, which is standard for enterprise platforms and also means the number that decides the deal is negotiable.

Our opinion

Cohere's strongest argument is not any single feature but the deployment story. Being able to run the same agent platform fully air-gapped is what banks, hospitals and government departments need before they let a model anywhere near their data, and it is the part of the enterprise market the hyperscalers find hardest to serve. The cost controls matter just as much, because agentic AI's dirty secret is that a looping agent can quietly spend a fortune, and organisation-wide caps are really a governance feature wearing a finance badge. What is missing is proof: Cohere names LG CNS and Bell Cyber, which is a start rather than a track record. North 2 is a credible answer to a real problem, and the next couple of quarters will show whether enterprises agree.