Centalion Group, one of the world’s largest independent commodity trading companies, has deployed an AI Treasury Agent that lets treasury professionals run facility and trade finance instrument workflows through natural language, with every control and approval step kept in place. Komgo, the digital trade finance platform provider whose Konsole product underpins the agent, announced the deployment together with Centalion.
Centalion has used Komgo’s Konsole for more than five years as the central platform for managing its off balance sheet instruments across its banking counterparties. Even so, the treasury team still had to move between systems to retrieve data, validate information and chase status updates. The companies said the bottleneck was no longer digitalisation but the way users interact with increasingly complex infrastructure.
Centalion’s answer was to build and govern the agent in its own cloud environment, combining a large language model with structured treasury workflows and facility specific business rules. A user can ask whether enough facility and fronting capacity exists to issue or amend instruments, and the agent interprets the request, retrieves the data, applies the facility controls and prepares the transaction for human approval.
The companies stressed that it is not a chatbot, since governance, validations and approvals are enforced at the orchestration layer and the user remains accountable for every decision. The link to Konsole runs through Komgo’s Model Context Protocol connector, so Centalion keeps control of its own AI and orchestration logic and needs no point to point integration project.
The first deployment covers Centalion’s off balance sheet instrument syndicated facility. According to the companies, issuance processing time has been cut by around 55% and amendment processing time by around 80%. Once the approach is extended across all of Centalion’s trade finance activity, they project it will free more than 60 working days of treasury capacity each year. Centalion is now extending the architecture to its other trade finance facilities, and the framework is designed to be replicated across other treasury and trade finance systems.
Olivier Thevenon, Treasury Manager at Centalion, said the team did not want another chatbot but a different way of working with technical infrastructure, in which the agent prepares a transaction in seconds and the decision stays with people.
“We didn’t want to just build another chatbot. We wanted to change the way our Treasury team interacts with technical infrastructure. The agent can understand an instruction, apply our rules, and prepare a transaction in seconds while the decision remains with the people in the Treasury team. For us, this deployment demonstrates the real potential of agentic AI in a commodity trader’s Treasury department: removing friction without removing accountability,” Thevenon said.
Baptiste Audren, Chief Revenue Officer at Komgo, said the deployment shows corporate treasury teams can build and govern their own AI on top of structured trade finance infrastructure instead of replacing it.
“Through this deployment, Centalion and Komgo are demonstrating how corporate treasury teams can build and govern their own AI capabilities on top of trusted, structured trade finance infrastructure, rather than replacing it,” Audren said.
For treasury teams that already work on digital platforms, the gain is in no longer switching between systems for routine checks, while approvals stay with people. For Komgo, a customer that builds its own AI on top of Konsole shows how the platform’s structured data can serve as a base for corporate AI, though the companies have not named the banks involved or given dates for the wider rollout.
Digital Trade Outlook’s View: Who Built the Agent Matters
What stands out is who built the agent. Centalion built it and runs it in its own cloud, and it connects to Konsole through a standard connector, while Komgo provides the structured trade finance data underneath. That makes Komgo the foundation for a customer’s own AI, not the AI vendor itself. The savings so far come from a single syndicated facility, and the 60 working days a year is only a projection, so the real test is whether the results hold as the agent takes on more instruments.