AI that explains your records. People who decide.
Humboldt, the AI analyst, reads the findings in your records, explains each one in business words and suggests what to do next. Custom agents, now in design, will each take on one recurring job. Neither changes your records on its own.
Humboldt runs on Claude by Anthropic.
People approve every change
The AI prepares and suggests. Nothing reaches your records until a person with the right to approve confirms it.
Humboldt, the AI analyst
The records a company keeps can contradict each other in ways nobody notices by hand: more coffee out of a huller than went in, a certificate issued after the goods left, a large plot recorded as a single point. Findings show them before a buyer, an auditor or an authority does.
Controls decide
Versioned control packs for EUDR, FSMA 204, certification schemes and your own policy open a finding for each inconsistency or missing piece of evidence. The same records always give the same findings.
Humboldt explains, groups and suggests
On request, Humboldt explains each finding in business words, groups findings that share one cause into cases, ranks them and suggests up to three next steps. Every sentence cites the records.
People decide
A person acknowledges, resolves or dismisses each finding, and turns a suggestion into a draft with one click. Humboldt never creates or sends anything itself.
What Humboldt never does
- It never says that a lot, a product or a company complies. Its text is checked for verdicts before it is stored.
- It never reads another company’s records, and works under the same access rules as the person who asked.
- An AI check is never evidence. A finding points at records; it does not replace them.
Custom agents
A digital colleague of your own for one recurring job: reading the certificates suppliers e-mail you, making sure every EU-bound lot has its EUDR evidence before it leaves, chasing suppliers whose certificates expire.
In designCustom agents are in design: nothing described here is built yet.
A playbook, not a prompt
You write the job the way you would brief a new clerk: the goal, the steps, your own rules and exceptions. It is drawn as a graph you can read and change.
You choose when it starts
An event in your records, such as a certificate about to expire, a schedule, an e-mail to its own mailbox, or a person pressing Run.
Only the tools you grant
A small set: read records, prepare drafts and checks, write messages you review, propose changes. Approving is never one of them.
Tested before it goes live
Scenario tests and a fixed set of adversarial cases must pass, then at least seven days on trial where nothing leaves and every would-be action is shown.
Measured, version by version
How often people accept its work unchanged, the minutes it saves against a baseline you declare, and what it costs.
Never alone on your records
It reads, matches, chases and drafts. People approve anything that changes the records.
The rules every agent follows
Every agent on the platform works under the same rules: Humboldt, the assistant, custom agents and outside agents connected to your records.
Agents propose, people approve
An agent can read and prepare proposals and drafts. It never approves, publishes, sends or signs, and never says an action happened before a person approved it.
Records are data, never instructions
Text from other companies, e-mails and documents is quoted, never followed. Evidence is quoted at its level and never upgraded.
One company per credential
An agent works inside one company and one environment, with the narrowest access that does the job. Another company’s records stay out of reach.
See it with your own lots.
Tell us what you trade and where it goes. We will show you the platform with your products, your units and your rules.