AI tools can be brilliant, until they start whispering your secrets into some model’s training data. That’s been the holdup for a lot of big organizations: they want the efficiency of AI, but don’t want to risk trade secrets, customer data, or sensitive internal info leaking into the digital ether. Enter Cohere’s new AI agent platform, “North.”
Built for the super-cautious (think banks, governments, health systems), North promises total control, with private, behind-the-firewall deployment. Translation? Your data stays yours. No peeking. No training fodder. No cloud eavesdropping.
“LLMs are only as good as the data they have access to,” said Cohere co-founder Nick Frosst. “So let’s put them right where the useful data actually lives—inside your infrastructure.”
Unlike other AI platforms that rely on big cloud names like AWS or Azure, North can run locally on your servers, hybrid clouds, VPCs, even on a GPU stuffed into a closet, if that’s your vibe. It’s designed to work on as few as two GPUs and still be useful.
Related: Google DeepMind Unveils Genie 3: A Breakthrough World Model for Training General AI Agents
But Cohere isn’t just selling digital fortresses. North does things, like answering support questions, summarizing meetings, creating marketing content, building tables and slides, or digging up internal insights. Everything it produces comes with citations and reasoning chains, so employees can audit its work instead of blindly trusting a black box.
The tech behind it? North is powered by Command, Cohere’s family of enterprise-tuned LLMs, and Compass, their multimodal search stack. The model used is specifically trained for what Cohere calls “enterprise reasoning,” meaning it doesn’t just spit out answers; it does real work.
It also connects with common workplace tools like Gmail, Slack, Outlook, Salesforce, and more, and can hook into Model Context Protocol (MCP) servers for domain-specific data. Of course, it is not just about where it runs, but how secure it stays. North supports granular access controls, agent autonomy settings, ongoing red teaming, third-party security tests, and checks every international compliance box, GDPR, SOC-2, ISO 27001, and friends.
Cohere, now valued at $5.5 billion after raising nearly $1B in funding, has already tested North with big names like RBC, Dell, LG, Palantir, and Ensemble Health Partners. They also recently acquired Ottogrid, a company that builds high-level enterprise research tools, meaning North just got a research brain boost.
“The more you use it,” Frosst said, “the more you go from ‘this helps me’ to ‘this handles it for me.’”
Cohere isn’t just selling an AI assistant. They are selling digital confidence. Want AI without the existential fear of leaking your data to a black box in Silicon Valley? North might just be the first enterprise AI agent that earns your trust before it earns your clicks.
Are we finally seeing AI that enterprises can safely say yes to?
