How is North 2 different from a chatbot?

A chatbot answers one prompt and stops. A North 2 agent takes a goal, plans steps, uses tools through connectors, checks its work, and escalates to a human when its autonomy policy requires approval.

Can North 2 run fully offline?

Yes, according to the video: Cohere supports deployment in the cloud, a VPC, hybrid, on-premises, or air-gapped with no internet connection. That capability addresses sensitive-data handling, though security certifications and controls should be verified directly with Cohere before any regulated deployment.

Does North 2 work only with Cohere models?

No. The platform supports Cohere models and lets companies bring their own, which reduces dependence on a single provider.

How does North 2 control agent spending?

Admins set token limits, rate limits, user quotas, and company-wide caps, with alerts when limits are hit. Usage is visible per user and per agent, so runaway consumption surfaces before it becomes a bill.

How to Use Cohere North 2 in Your Company

Cohere North 2 launched, per the video coverage, around October 5, 2026, as an enterprise AI agent platform. Its main parts are an agent harness, reusable skills, shared libraries, persistent memory, automations, connectors to work tools, and admin controls over security and spending.

What Is Cohere North 2?

Cohere North 2 is an enterprise AI agent platform: a full-stack system where companies connect agents to private data, let them run multi-step tasks, and keep control over security, permissions, and cost. The presenter calls it a "super agent," but the vendor positions it as a platform, and the distinction matters: the value is the surrounding harness, skills, memory, and admin layer, not one model.

Cohere, the Toronto-based enterprise AI company behind the Command models, says a year of deploying the earlier North product in high-stakes environments shaped the upgrade. The video dates the North 2 release to October 5, 2026; treat that date as the video's claim and check Cohere's North page for the official announcement.

The audience is large organizations with private data, not individual consumers. That framing explains most of the feature list below: everything exists so an agent can act without either a human babysitting every step or IT losing track of what it did.

How the Agent Harness Works

The agent harness is the engine that plans and executes steps instead of answering once. The transcript contrasts the two modes directly: a chatbot takes your question and replies; an agent takes a goal, builds a plan, calls tools, completes many steps, checks its own work, and asks a human to approve when policy requires it.

A workable example from the video: research what a community wants from AI automation, then draft a plan for five videos on the topic. That is a job with research, synthesis, and deliverables, which is the workload class the harness is built for.

Skills and Libraries: Reuse Across the Company

Skills solve the problem of repeating setup work. A skill is a saved procedure for one task, such as a weekly report that pulls data, flags problems, and writes a summary. Once built, every agent and every team member can invoke it, so nobody rebuilds the same workflow or skips a step.

Libraries are the knowledge counterpart. They hold policies, documents, brand rules, templates, and research in shared places that agents can read. The practical effect the transcript describes is consistency: agents stop improvising tone because every one of them draws on the same source material, and employees stop pasting the same context into every prompt.

Memory, Applications, and Automations

Memory lets agents keep context across sessions, so you explain your business once and stop repeating yourself. Note the caveat the presenter himself states: Cohere has not published exact figures on how much context North 2 retains, so any capacity number here would be a guess.

Two build features sit alongside memory. Applications generate documents, dashboards, decks, tables, charts, and small apps from plain-language requests. Automations are shareable workflows with ready-made templates; you can describe one in plain words or assemble it with drag-and-drop, watch it run in real time, and connect it to existing tools. The video's example: a new lead asks a question, the agent researches it, drafts a reply, and routes it to a person for approval before sending.

Which Tools Does North 2 Connect To?

Connectors let agents act inside the software a company already runs rather than just talk about it. The video lists these integrations, and Cohere says more data connections are planned but not yet available, so treat the planned ones as roadmap, not shipping features.

The connectors named in the review, each linked to its official site:

Slack, for team chat
SharePoint and OneDrive, for files
Outlook and Exchange, for email
Jira and Linear, for work tracking
Notion, for docs and projects
GitHub, for code

A combined task the video sketches: find every question about a topic across team chat and notes, then write a post showing how the company answers them. That only works because the connector layer reaches both systems.

Security, Autonomy Policies, and Usage Limits

Security is where North 2 differs most from consumer agent demos. Deployment is flexible: cloud, VPC, hybrid, on your own servers, or air-gapped with no internet connection at all. Per the video, Cohere states North holds SOC 2 Type 2, ISO 27001, and ISO 42001 certifications, and runs guardrails, detailed logs, security testing, red teaming, and third-party scans. Verify the current certification status on Cohere's trust page before relying on it in a compliance decision.

Autonomy policies decide what an agent may do alone, what is off limits, and what requires human sign-off. The flow is simple: the agent wants to act, a policy checks it, and borderline actions wait for a person. This is a governance design, not an afterthought bolted onto a chatbot.

Usage limits close the cost hole. Because an agent can fire off thousands of model calls with no one approving each one, North's admin console shows usage per user and per agent and lets admins set roles, permissions, model access, token limits, rate limits, user quotas, company-wide caps, and alerts. The pattern: agent spends, hits a limit, alert fires, admin intervenes.

Finally, the platform is not locked to one model. Companies can use Cohere models or bring their own, which the transcript frames as insurance against single-vendor lock-in.

Frequently Asked Questions

What is Cohere North 2? Cohere North 2 is an enterprise AI agent platform for building agents that connect to company data, run multi-step tasks, and operate under admin-defined security and usage rules. Cohere positions it for large organizations with private data, and the review coverage dates its release to early October 2026.

How is North 2 different from a chatbot? A chatbot answers one prompt and stops. A North 2 agent takes a goal, plans steps, uses tools through connectors, checks its work, and escalates to a human when its autonomy policy requires approval.

Can North 2 run fully offline? Yes, according to the video: Cohere supports deployment in the cloud, a VPC, hybrid, on-premises, or air-gapped with no internet connection. That capability addresses sensitive-data handling, though security certifications and controls should be verified directly with Cohere before any regulated deployment.

Does North 2 work only with Cohere models? No. The platform supports Cohere models and lets companies bring their own, which reduces dependence on a single provider.

How does North 2 control agent spending? Admins set token limits, rate limits, user quotas, and company-wide caps, with alerts when limits are hit. Usage is visible per user and per agent, so runaway consumption surfaces before it becomes a bill.

Source video