The rule everyone knows and nobody has tested
“We only ship once the delivery acceptance is signed.” “Discounts above 15% need CFO approval.” “A department can’t report to its own sub-team.” Every company runs on rules like these. They live in a policy doc, in the controller’s head, and in a pile of exceptions only one person remembers.
Trouble shows up at the edges. Is 15% “above 15%”? What if the acceptance is signed but payment hasn’t cleared and the customer is a regular? With ten rules in play, nobody sees the contradictions until a disputed shipment goes out. And when an AI assistant is making those calls, it can be confidently wrong.
LOGUS works on one principle: logic shouldn’t be guessed, it should be computed. A rule becomes a process you can run, test against every input, and explain line by line.
How LOGUS connects to the rest of PRANUS
LOGUS is the shared logic engine for the whole platform, so its rules apply where the work actually happens.
- Org structure. TEAMUS checks the org chart through LOGUS for circular reporting, where a department ends up reporting to itself.
- Tasks. In TASKUS, logic runs right on the task canvas: an approval flow sits next to the tasks it governs and executes on live data.
- AI employees. Agents in AGENTUS send “allowed or not”, “how much”, and “what status comes next” questions to LOGUS and get an answer with a confidence level attached.
- Manufacturing. The product configurator in OPTIMUS uses LOGUS to decide which combinations of options are valid.
- Personal assistant. Assistentus includes a LOGUS helper, so you can describe a rule by voice or text and keep the schema as a note.
What the AI does, and what the engine proves
The split matters. AI helps you express the rule: it turns your description into a schema, drafts test scenarios from your wording, and suggests precise edits. The checking is done by a solver and by exhaustive enumeration, not by a language model.
That’s why every LOGUS answer carries a label:
- Proven: verified by a solver or by checking every possible input;
- Computed twice: calculated in two passes over the same evaluator;
- Computed: calculated once;
- Estimated: no full check was possible, and LOGUS says so;
- Mismatch: results disagree and a person should look.
When a rule is wrong, you don’t get a generic error. You get a counterexample: “with an order of $15,000 and customer status regular, shipping without signed acceptance is allowed, but it shouldn’t be.” When rules conflict, LOGUS shows the smallest set of rules responsible, so the fix is obvious.
Example: a discount policy in five minutes
Say your head of sales writes: “Reps can approve discounts up to 10%. Between 10% and 20%, the sales manager approves. Above 20%, the CFO. Repeat customers get an extra 5% with no approval.”
LOGUS turns that into a process with an input form: discount size, customer status, approver. It drafts test scenarios from your wording and probes the boundaries: 10%, 10.01%, 20%, 25% for a repeat customer. One of them reveals that the “extra 5%” clause lets a rep give a repeat customer 14.9% without a manager, which isn’t what anyone intended. You find out before the margin is gone, not at the quarterly review.
Once fixed, the rule is marked Proven, and people and AI agents apply it the same way. Earlier versions stay in the log, so you can always see what changed and when.
Getting started
Most teams get their first rules during the PRANUS rollout: the org structure the onboarding agent builds is checked by LOGUS right away. From there you add rules as you need them, test them at the boundaries, and connect them to tasks or AI agents. No separate process-modeling project required.
Who it’s for
- Owners and executives who want company rules applied the same way every time, not from memory.
- Finance and legal teams managing limits, approvals, shipping conditions, and discount policies that can’t be broken.
- Business analysts and operations leads who model processes and want dead ends caught before launch.
- Teams handing decisions to AI agents who want those agents to compute answers instead of inventing them.
Want to see one of your own rules turned into a verifiable process? Book a PRANUS demo.