GuluStock and AutoServa · GuluStock and AutoServa: User Guide
This AutoServa glossary explains the words used in a working session, in plain language. AutoServa is VYROX's service for running an agentic AI operator on a business queue. It is written for business owners, managers and risk or compliance colleagues.
| Term | Meaning |
|---|---|
| Operator | The tuned AI that works one of your queues, in the position of an operations staff member between your business and its systems |
| Decision Framework | A written structure of inputs, thresholds, constraints, approvals and escalation. The main output of a decision audit |
| Threshold | The value at which an answer changes, such as a tolerance percentage or a spending limit |
| Escalation Rule | When a case stops being routine and must go to a person. Set per workflow, not guessed by the AI |
| Precedent | What you decided last time, kept searchable |
| Queue | The stream of repeating work one engagement targets, one queue at a time |
| Shadow Running | The operator works live cases but nothing it produces is acted on, so accuracy can be measured first |
| Disagreement Log | Every case where the operator and a person differed, reviewed weekly |
| Term | Meaning |
|---|---|
| Workflow Guard | Checks every action against approval chains, spending ceilings, access scope, segregation of duties, retention rules and regulatory obligations before it runs |
| Evidence Trail | What was read, which rule applied, which threshold was hit, which precedent it followed and who approved |
| Human in the Loop | A person approves before anything irreversible is final |
| Segregation of Duties | The preparer and the approver are different parties |
| Least Privilege | Only the access a role needs. The operator starts read-only |
| Data Residency | A rule that data stays in a country or region |
| Air-Gapped | A system with no outside network connection, the strictest local deployment |
| Term | Meaning |
|---|---|
| Agentic AI | AI that plans steps and acts toward a goal, rather than answering one question at a time |
| Frontier Model | The most capable model a provider offers now |
| Open-Weight Model | A model you can run on your own hardware |
| Fine-Tuning | Extra training on your documents, decisions and vocabulary |
| RAG | Retrieval-augmented generation: fetch the passages a case needs from your documents before the model answers |
| Embedding | A numeric form of text that finds similar meaning, not only matching words |
| Knowledge Graph | Your framework held as connected entities and rules, such as vendor, contract, clause and precedent |
| Reasoning Trace | The path the operator took to an answer, kept with the evidence |
| Hallucination | A confident statement that is false |
A chatbot answers questions. The operator works a queue, acts inside your systems under your rules, and shows evidence.
No. Nothing the operator produces is acted on during shadow running.