Cohere’s North 2 Puts AI Agents on a Budget. Toronto Bets on Boring

The line in Cohere’s North 2 launch post that matters most to a chief financial officer is a dull one. Buried in a list of admin features, the Toronto company says customers can now “monitor and control token spend and limiting by defining consumption tiers based on requests and token rates for users and groups.”

Translated: an AI agent at your company can now hit a spending ceiling and stop, instead of quietly running up a bill nobody sees until the invoice lands.

Cohere released North 2 on Oct. 5, calling it the company’s “biggest upgrade yet” to the agent platform it made generally available in August 2025. The flashy parts are there. The pitch, though, is built on the boring parts: budgets, permissions, audit logs and a server that never has to touch the public internet.

What Cohere North 2 adds for AI agents

The core change is a rebuilt orchestration layer that breaks multistep jobs into pieces and pulls a human in where a step needs approval, SiliconANGLE reports. Around it, Cohere lists reusable “skills,” shared knowledge libraries, memory that carries context across sessions, generated decks and dashboards, prebuilt automation templates with a drag-and-drop workflow builder, and connectors to Slack, SharePoint, OneDrive, Outlook, Exchange, Jira, Linear, Notion and GitHub, with financial data providers still to come.

Security sits underneath. North 2 screens for personal information and prompt-injection attempts, restricts each agent to authorized actions, and runs on-premises, in a customer’s cloud or fully air-gapped. Cohere cites SOC 2 Type 2, ISO 27001 and ISO 42001 certifications.

It is also model-agnostic. The Logic reports customers can plug in models from Anthropic, OpenAI, Google, Alibaba, DeepSeek and Z.ai. That is an unusual move for a company that sells its own models, and Paul Teyssier, Cohere’s vice-president of product for AI, framed it as a jab at rivals: “A whole bunch of players [are] trying to take the customers deeper and deeper into their ecosystem, with no way out. We are intentionally taking the other approach.”

Why token spending caps lead the pitch

Teyssier told The Logic that cost control is “probably the number one rising concern for AI leaders.” There is data behind the worry. In a July survey from VentureBeat’s VB Pulse tracker, 21% of enterprises said they only monitor agent spending reactively, with no programmatic way to stop an agent before the bill arrives. The same survey had Microsoft as the leading agent platform at 41%, with Anthropic’s Claude at 28%.

“The platform that’s secure enough wasn’t smart enough; the one that’s smart enough wasn’t controllable enough; the one you could govern became too expensive to run at scale.” — Cohere, North 2 launch post

Two things are missing. Cohere has not published North 2 pricing, telling VentureBeat it is “priced based on the scale and complexity of a customer’s deployment.” And on memory, analyst Rajesh Beri points out that Cohere has not said where agent memories are stored, how long they are kept or how they get deleted. For a product sold on control, that is a gap a compliance team will notice first.

Bell, LG CNS and Canada’s sovereign AI bet

The launch quotes lean on customers who care about where data lives. “North lets us bring agentic AI into security operations with control, privacy, and sovereignty built in from the start,” said Jawed Ahmad, chief technology and AI officer at Bell Cyber. Yohan Jin, head of the AI Center at South Korea’s LG CNS, called North “a secure, practical foundation for scaling intelligent automation.” Nvidia’s Kari Briski said Blackwell and Hopper GPUs help North “deliver more tokens for less.” Radical Ventures, an early Cohere investor, built an automation that prepares briefing materials for founder calls, per SiliconANGLE.

Ottawa is part of this story too. The federal government committed up to CA$240 million in March 2025 toward a Canadian AI data centre for Cohere, then signed a non-binding agreement in August 2025 to explore its tools in the public service. The Logic lists Canada’s innovation department among North clients, alongside RBC and Bell. Radical co-founder Jordan Jacobs now sits on Carney’s new National Council on Artificial Intelligence.

The same day as North 2, Cohere announced a global alliance with PwC that launches first in Canada. Its Embed 5 retrieval models shipped Sept. 30, and on Sept. 16 it signed an agreement to combine with German developer Aleph Alpha, with headquarters in Toronto and Berlin, pending regulatory approval.

Cohere’s IPO case and what it means for you

All of this feeds an IPO story. Cohere hit US$240 million (about CA$334 million) in annual recurring revenue in 2025, according to an investor memo first reported by CNBC, and CEO Aidan Gomez has said a listing could come “soon.” Our read: selling governance to banks, telcos and governments is how a company with no consumer chatbot hit shows public-market investors durable revenue. (Our guide to reading an AI company’s S-1 covers which numbers to check if a filing appears.)

If you run IT at a Canadian organization weighing North 2, ask three things before signing: whether caps can hard-stop an individual agent mid-task, where agent memory is stored and how it is purged, and what the price looks like at your token volume. None of those answers are on Cohere’s site yet.

Teyssier’s closing argument to The Logic was that customers will “prefer our models based on hard facts.” Cohere told VentureBeat that application releases are harder to benchmark than models, and that it is still working on data it can share.