Scattered data
Contracts, shipments, payments and positions are scattered across emails, spreadsheets and tools that do not communicate.
AI agents built for commodity trading.
We design and deploy AI agents for trading houses and trading desks: document control, logistics tracking, finance, risk and compliance. Every solution is built to measure and supported over time.
01 · The problem
Trading still runs on email, Excel and PDF.
A physical trade involves dozens of documents and exchanges between front office, operations, logistics and finance. Most of that work is still manual: re-keying, tracking, matching, proofreading.
Major houses have entire teams to absorb it. Mid-sized houses and specialised desks rarely do. This is where AI agents, built with genuine knowledge of the trade, come in.
Manual tracking
Every document, vessel and due date is tracked by hand. One oversight can hold up a payment or cost demurrage.
Overloaded teams
Operators spend their time re-keying data instead of analysing and managing counterparties.
02 · Areas of expertise
From front to back office, wherever time is lost.
Every desk is different. We start from your processes to target repetitive, time-consuming tasks: the ones an agent performs faster and more consistently, under your team's supervision.
-
01
Documents & back office In production
Reading and checking contracts, confirmations, letters of credit, bills of lading and certificates. Generating confirmations and invoices.
-
02
Automated logistics tracking
Vessel and container tracking, estimated arrival dates, nominations, delay alerts, demurrage calculations and checks on weight or quality discrepancies.
-
03
Finance & treasury
Bank reconciliations, monitoring of collections and payment schedules, cash forecasts, tracking of letters of credit and currency exposure.
-
04
Risk & reporting
Position book consolidation, price exposure monitoring, daily P&L reporting and limit breach alerts.
-
05
Market intelligence
Monitoring of official reports, trade press and market data, summarised for the desk before the open.
-
06
Compliance
Counterparty identification, sanctions screening, KYC file preparation and a complete audit trail.
-
07
Commercial
Reading customer requests, preparing quotes, structuring supplier catalogues and tracking counterparty correspondence.
Workflow example
Checking a letter of credit presentation.
The agent reads each document, checks it against the letter of credit requirements and flags discrepancies before presentation to the bank. The operator approves. This workflow is in production; it shows the logic we apply to logistics, treasury or reporting.
03 · AI agents
Not a chatbot. An equipped operator.
Each agent has a defined role, works with the desk's tools, applies its rules and stops wherever a human decision is required. An orchestrator coordinates the whole and picks the right model for each task.
Inputs
- Email & messaging
- Contracts & documents
- Logistics data
- Banks & ERP
- Market data
The desk, augmented
Orchestrator
Multi-model- Documents agent
- Logistics agent
- Treasury agent
- Risk agent
- Compliance agent
- Market agent
- Reasoning models
- Fast models
- Self-hosted models
Desk memory · business rules · trade history
Human approval
Outputs
- Checked & generated documents
- Logistics alerts
- Position & cash
- Briefings & reports
- Audit trail
Supervised
No binding action without an operator's approval. The agent prepares, the human decides.
Traceable
Every read, every calculation and every proposal is logged and can be audited.
Integrated
Agents work inside existing tools. Teams keep their habits.
04 · Support
Bespoke solutions, at your pace.
No off-the-shelf product and no eighteen-month programme. We start from your needs, put a first workflow into production, measure results on your real operations, then expand.
- 01
Desk audit
Mapping of processes, time spent and friction points, together with the operators.
DeliverablePrioritised roadmap - 02
Design
Choice of use cases, architecture, models, business rules, security and quantified success criteria.
DeliverableApproved specification - 03
Deployment
Pilot on a live workflow, before-and-after measurement, then gradual roll-out to production.
DeliverableAgent in production - 04
Ongoing support
Team training, monitoring of performance and costs, improvements and extension to new workflows.
DeliverableSelf-sufficient team
05 · Technology & trust
The right model, for every task.
Our architectures are multi-model: each task goes to the most suitable model given the accuracy required, volume, latency and cost. The configuration is specific to each desk.
TaskMain criterionModel type
Checking a letter of creditAccuracy · reasoningAdvanced reasoning model
Shipment tracking and alertsVolume · responsivenessFast, low-cost model
Treasury reconciliationsReliability · structureData-specialised model
Highly sensitive dataConfidentialitySelf-hosted model
Models
Anthropic's Claude, open-source models and specialised models, combined and arbitrated task by task. No dependency on a single provider.
Integration
Email, spreadsheets, document storage, banks, ERP and ETRM through APIs. Agents come to the desk's tools, not the other way round.
Data
Providers selected for their contractual guarantees: no training on the data sent. Hosting in Europe or on the client's own infrastructure.
Control
Human approval on every binding action, full logging, role-based access, cost tracking per workflow. Designed with GDPR in mind.
Prices, commitments and commercial decisions remain human. Agents serve the speed and accuracy of execution.
06 · FAQ
AI agents and trading, the essentials.
What is an AI agent for commodity trading?
A program that performs a specific desk task: reading a document, tracking a shipment, reconciling payments, preparing a report. It works with existing tools, applies the desk's rules and submits every binding action for human approval.
Which processes can be automated in a trading house?
Checking commercial and transport documents, logistics tracking, finance and treasury, position reporting, market monitoring, counterparty compliance and quote preparation.
Which AI models do you use?
Our architectures are multi-model: Anthropic's Claude, open-source models and specialised models, chosen task by task based on accuracy, volume, latency and cost. Every configuration is specific to the desk.
Does our data stay confidential?
Yes. We select providers that do not train their models on the data sent, offer hosting in Europe or on your own infrastructure, and self-hosted models for the most sensitive data.
Do we need to change our tools?
No. Agents integrate with your existing email, spreadsheets, document storage, ERP and ETRM.
How do we get started?
With a first 30-minute call, then a desk audit that identifies priority workflows. A first agent is piloted on a live workflow before any extension.
Contact
Let's talk about your project.
Trading or artificial intelligence: write to us and we will reply within 24 business hours.