Cohere North

• Speeding up contract review and redlining and checking documents for compliance across large volumes of legal material
• Building a scheduled automation that pulls CRM, Slack and calendar activity each morning and posts a briefing for managers
• Turning plain-language questions into SQL against a governed data warehouse and returning the answer to analysts
• Preparing for sales calls by combining CRM opportunities, email and Slack history, notes and market research into a presentation
• Rolling out shared agents to a regulated workforce while tracking token spend per user and agent and capping it by user group and organization
• Running agentic AI over sensitive data in an air-gapped or on-premises environment where cloud assistants are not allowed

North competes less on the assistant itself than on where it runs and who controls it. A model-agnostic agent workspace that can sit in your VPC, on-premises or air-gapped, with token limits per user and group, spend visible down to each agent, and approval steps in workflows, answers the questions that stall agent rollouts in banks, hospitals, telecoms and governments. The cost is opacity: no public price, few public docs on how agents and automations are built, and some headline connectors still planned. Shortlist it if deployment location and governance decide your choice. If your staff already live in Microsoft 365 or ChatGPT and a hosted service passes your security review, the suite-native options will be quicker to roll out.

About Cohere North

North is Cohere's agentic platform for knowledge work, separate from its Command, Embed and Rerank models and from Compass, its search product. Cohere announced North on January 9, 2025, and on October 5, 2026 released North 2, which it calls its biggest upgrade. Employees use a chat workspace to search and reason over company data, create summaries, tables and charts, and prototype decks, dashboards, documents and lightweight apps. They can also build agents and automations from prompts and share them across the organization. North 2 rebuilt the orchestration layer that runs these multi-step agents, and added skills (reusable capabilities agents can call), shared libraries, and memory that carries context across sessions. Automations, launched in July 2026, turn a plain-language description into a workflow with scheduled runs, loops and branching, a model choice for each step, a Plan mode to review the approach before building, versioning, testing before publishing, and steps where a colleague must approve; North 2 added templates and a drag-and-drop visual builder. Autonomy policies limit agents to actions they are allowed to take and send critical decisions to a person. North connects to Slack, SharePoint, OneDrive, Outlook, Exchange, Jira, Linear, Notion and GitHub, and to your own tools through MCP. Cohere publishes an MIT-licensed Python SDK for building MCP servers that use North's authentication. North is model-agnostic: you can use Cohere's models or bring other LLMs, and admins control which models each user or group can use. What sets it apart from suite-bound assistants is where it runs and how it is governed. It can be deployed self-hosted, in your VPC, hybrid, on-premises or air-gapped. Cohere's docs also describe running the North application in your own network while sending model inference to a Cohere-managed Model Vault. North Admin shows usage and token spend down to each user and agent, sets consumption tiers for users and groups and org-wide caps, and can alert teams at thresholds before limits are hit. The tradeoffs: pricing is custom and there is no self-serve sign-up, there are no public product docs beyond deployment notes and the MCP SDK, and getting value from it depends on the integration work an enterprise rollout needs.