Cohere launches North 2 enterprise AI agent platform with memory and cost controls

By Splitfeed EditorialLast updated Oct 6, 2026, 07:10 UTC

On October 5, 2026, Cohere released North 2, an updated enterprise AI agent platform built around four central capabilities: persistent memory that retains context across sessions, token spending caps that let organizations set hard limits on compute costs, access control lists that restrict which users or systems can trigger specific agents, and a rebuilt orchestration layer that handles multi-step workflows autonomously. The platform also accepts models from providers other than Cohere, and it offers sovereign deployment options aimed at organizations with strict data residency or compliance requirements.

The release matters because it addresses two persistent friction points that have slowed enterprise adoption of AI agents: unpredictable cost and unpredictable behavior. Token spending caps are a direct response to complaints that agentic systems can run up compute bills without clear stopping points, and the access control architecture attempts to give IT and compliance teams the governance hooks they need before approving deployments at scale. Supporting third-party models widens the addressable market and positions North 2 as a neutral control layer rather than a Cohere-only ecosystem, which is a meaningful strategic choice given that most large enterprises already operate across multiple model providers. What remains genuinely in dispute is whether soft controls like spending caps and access lists are sufficient to satisfy regulators and risk officers, or whether the autonomous, multi-step nature of the agents introduces liability questions that platform-level governance cannot fully resolve.

No reactions from the Pro-AI, Anti-AI, or Middle Ground camps have been published yet. Typically, the Pro-AI camp would treat features like token caps and access controls as evidence that enterprise-grade AI is maturing responsibly, arguing that governance tooling makes broader deployment safer and more justified. The Anti-AI camp would likely question whether configurable controls inside a vendor platform constitute meaningful oversight, and would push back on the framing of autonomous agents as ready for sensitive enterprise environments without stronger external accountability. The Middle Ground camp would probably welcome the cost and access controls as necessary steps while cautioning that memory persistence and autonomous orchestration raise data-handling and decision-audit questions that enterprises should scrutinize carefully before committing.

The clearest near-term signal to watch is whether large regulated-sector customers in finance, healthcare, or government publicly adopt North 2, and whether any compliance or audit frameworks specifically address persistent-memory agent deployments. Announcements from Cohere about enterprise contracts or certifications in those sectors would help clarify whether the governance features meet real institutional thresholds or serve mainly as marketing positioning.