Glossary

An Agent is the platform's fundamental intelligent unit — a conversational entity with identity, knowledge, capability and boundaries, representing a merchant, industry, organization or lifestyle scenario. Four parts define what it may say, must not say and can do. The platform recognizes four kinds: Industry Agent (Type A), Local Service Agent (Type B), Organization Agent (Type C) and Enterprise Solution Agent (Type D).

The Brain is an agent's reasoning core: it interprets the question, selects knowledge, forms the judgment and decides the answer structure. The platform follows a single-brain principle — within one product domain all intelligence lives in one place, never scattered across UI components.

Case Analysis is the structured judgment an agent makes about a user's specific situation rather than a generic explainer. It typically contains a conclusion, its basis, the risks and the next actions — the dividing line between professional dialogue and encyclopedic answers.

A Fact Pack holds deterministic facts — limits, rates, deadlines, thresholds, form numbers — where ambiguity is unacceptable. Fact questions take a deterministic path: the same question must always yield the same answer, and when jurisdiction or source data is missing the agent refuses explicitly instead of guessing.

Knowledge is the factual and procedural material an agent relies on: definitions, rules, procedures, thresholds and timelines. The platform separates public knowledge from real lived experience — the latter is its distinctive asset. Knowledge enters an approved skill pack only through human review; automatic promotion is not permitted.

A Knowledge Pack bundles knowledge around one subject — a merchant pack, an industry pack, an organization pack. Packs are deliverable, versioned assets submitted at onboarding and maintained in the workspace.

Memory is user-related information retained across turns or sessions — city, household situation, prior concerns, confirmed preferences. It reduces repeated questioning and improves continuity; anything the user states explicitly outranks anything the system infers.

Reasoning is the judgment process an agent completes before answering: classifying the question, drawing on facts and experience, weighing risk and reaching a conclusion. The platform's rule is verdict-first — even with incomplete context, give a usable judgment from sensible defaults, then ask at most one key follow-up.

A Skill is a professional action or answering capability an agent can perform — eligibility assessment, document checklists, timeline planning, risk flagging. Skills are composable and reusable: industries share generic skills while keeping their own specialized ones.

A Skill Pack is the approved, agent-callable combination of skills and knowledge — the formal output layer of the knowledge system. Only human-reviewed candidates become skill packs, and high-risk industries (legal, immigration, tax, insurance, medical, financial) always require review.

Synthetic Q&A are question-answer pairs generated from already-verified facts to broaden coverage of how users may phrase the same question. They may only derive from verified facts and must not introduce new ones — they expand phrasing, never the knowledge boundary.

A Tool is an external function an agent can call during a conversation — retrieval, calculation, form submission, booking writes, report generation. Tool calls must be explicit, traceable and compliant with their write contracts (for example, leads may only be written through the single canonical path).

Website Facts are structured facts extracted and verified from an official site: address, phone, email, services, fees, office hours, application procedures. They are the first-hand basis for merchant and organization agents and outrank generic model knowledge.