Product3 publishers3 min readPublished
Cohere pitches its North 2 agent platform to the administrator who has to cap the token bill
Cohere's North 2 agent platform lets administrators track token use down to each user and agent and cap spending across the whole company. Cohere is selling to whoever answers for agent cost and risk, and the launch report backs that with three customer accounts and no prices.
The Product Desk · Product desk
Drafted by a language model from the sources cited here and checked against its claim ledger before publication. How we use AISend a correction
What happened
- North 2's rebuilt orchestration works through multistep tasks on its own and pulls in a human for any step that needs sign-off.
- North has been on general release since August 2025. Cohere said the upgrade was shaped by a year of big production deployments in finance, government and other sectors.
- Customers can still deploy the platform on-premises or inside their own virtual private cloud, with fully air-gapped setups among the options.
- Connectors now cover Slack, GitHub, SharePoint and OneDrive, and Cohere plans connectors to PitchBook, S&P Global and FactSet.
- The release lands under three weeks after Cohere agreed to merge with Aleph Alpha, the German AI developer. The combined company will keep the Cohere name.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- decision Because teams can be assigned models from outside Cohere, a company choosing North commits to Cohere's console and the shared agents it builds there, and can keep changing the models underneath.
- cost Per-user tiers and company-wide caps give IT a standing job: setting token budgets team by team and answering the alert when one runs low.
- exposure Once a prompt-built agent is shared across the company and carries memory between sessions, any change by the team that owns it reaches every team that adopted it.
Before a founder call, the team at Radical Ventures used to spend hours gathering basic context, partner Parasvil Patel said [11]. The firm is an early Cohere investor, and it now has an automation built on North that prepares the briefing the team reads before the call [11]. That account and two others, from Bell Cyber and LG CNS, are the only user evidence in the North 2 launch [17].
Cohere frames its pitch as a tradeoff. The company says enterprises deploying agents have accepted giving up one of security, capability or cost control to get the other two [13]. What it shipped against that is mostly an admin console. In North Admin, administrators can put users and groups on consumption tiers set by request and token rates, and can apply a company-wide cap on usage [3]. Alerts warn teams before a limit is reached [4]. Per-agent guardrails screen prompts and responses for personal data and attempted prompt injection, and autonomy policies keep each agent to the actions it is authorized to take [14].
The evidence supports the governance reading, with cost close behind. North 2 is model-agnostic, so the models an administrator assigns to a team can include ones the customer brings from outside Cohere [7]. If the model can be swapped, what Cohere is really asking a company to adopt is the console. It has not dropped the price-per-token argument, though. Cohere says its models produce more tokens per second per node on Nvidia's Blackwell and Hopper GPUs [19], and Kari Briski, Nvidia's vice president of generative AI, said that helps North 2 "deliver more tokens for less" [20].
Bell Cyber runs North inside its own security operations [9]. Jawed Ahmad, its chief technology and AI officer, said North lets the unit run agentic AI with its customers' control and sovereignty requirements "built in from the start" [9]. LG CNS runs North internally and brings it to customers across South Korea [10]. Radical's story is about hours saved. None of the three is about the quality of model output [17].
SiliconANGLE's report does not include pricing, retention or usage figures [12]. For now, the case for governance rests on three customer accounts [17].
For the administrator who has to roll this out, two facts about the company decide whether North 2's design fits. One is who pays for tokens: a central budget or each team. The other is how many teams would run the same agent. A central budget with many teams sharing agents is the case North 2 is built for. Per-user tracking and company-wide caps have a job there, and so do agents shared across the company [2][6]. Team budgets with one-off agents is the cell where the console is overhead and the choice comes back to model output. North 2 lets a customer bring that model anyway [7].
The two mixed cells need one more test before anyone signs. Name the person who gets the alert when an agent nears its cap, and the budget that person would raise [4].
What to watch
- Whether Cohere publishes prices for North 2's consumption tiers, or any retention and usage figures from its production deployments.
- Whether Aleph Alpha's models appear as options teams can be assigned inside North once the merger closes.
- Whether rival agent platforms add per-user and per-agent token caps to their own admin consoles.