Cohere Adds Memory and Company-Wide Usage Caps to North 2 AI Agents
The enterprise platform combines reusable workflows, model choice and private deployment. Its new controls do not yet come with published performance results or standard pricing.
Loading page…
The enterprise platform combines reusable workflows, model choice and private deployment. Its new controls do not yet come with published performance results or standard pricing.
Listen to this story
North 2 lets teams share agent workflows and carry context across sessions, while giving administrators controls to track and limit usage. Cohere says it supports prompt and response checks, human approval for critical decisions, and deployments from cloud to air-gapped infrastructure. Buyers still lack published performance benchmarks and standard pricing, so they can assess the platform’s controls more readily than its results or total cost. Connections to PitchBook, S&P Global and FactSet are planned, not confirmed as available.
North Admin tracks token consumption by user and agent, and Cohere says teams can segment and export usage data by model, user and department.
Customers can use Cohere’s models or models they already use; Cohere says model, orchestration, search and embedding components are interchangeable.
Available connectors include Slack, GitHub, SharePoint and OneDrive; financial-data connections are planned.
Cohere is giving enterprise AI agents more continuity—and administrators more ways to contain them. Introduced on October 5, 2026, North 2 adds persistent memory, reusable workflows and company-wide usage caps. But the promise of more manageable automation arrives without performance benchmarks or published standard pricing, according to VentureBeat.
North Admin, the platform’s management console, tracks activity and token consumption down to individual users and agents. Administrators can set quotas, request and token-rate limits, and caps across the organization. Alerts warn teams before those limits apply.
The console also sets roles, permissions and which models different employees can use. It integrates with existing identity and access management systems. Paul Teyssier, Cohere’s vice president of Product AI, told VentureBeat that usage can be separated by model, user and department, then exported into customers’ analytics systems.
Cohere says agent-level guardrails scan prompts and responses for personally identifiable information and detect prompt injection, attempts to redirect an agent through malicious instructions. Autonomy policies restrict agents to authorized actions and send critical decisions to a person.
We only turned those workflows on once we could see and control what the agents were doing,
Jacob Rideout, HiddenLayer CTO, in a statement supplied by Cohere to VentureBeat
North 2’s redesigned orchestration system—the software that coordinates an agent’s work—handles complex, multistep processes and brings people in for designated decisions. Employees can build agents and automations from simple prompts, then share them across the organization rather than rebuild the same workflow.
The update also generates slide decks, dashboards and lightweight apps from plain-language requests, SiliconANGLE reported. Recurring processes get prebuilt templates and a drag-and-drop workflow builder with real-time performance reporting. The connector library includes Slack, GitHub, SharePoint and OneDrive; connections to PitchBook, S&P Global and FactSet are planned, not confirmed as shipping.
North can run Cohere’s models or models a customer already uses. Teyssier said its model, orchestration, search and embedding components are interchangeable, a design Cohere positions as reducing vendor lock-in. Deployment options include the cloud, on-premises infrastructure, a customer’s virtual private cloud and fully air-gapped installations with no outside connection.
Existing North users include Bell Cyber, which uses it in security operations, and LG CNS, which runs it internally and offers it to customers in South Korea. These are examples of the platform’s use, rather than measured results for the new release.
Cohere says pricing depends on deployment scale and complexity, including infrastructure, usage, support and customization. Teyssier said application releases are harder to benchmark than models and that Cohere is working on shareable data. For now, North 2’s controls are specified more clearly than its measurable performance or total deployment cost.
Loading discussion...
Join the conversation
Explain when finishing a task should outweigh a strict budget.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.