Glossary | Smart Agents

One-line reminders of the terms used throughout these manuals. For the full story with examples, read the AI concepts primer.

Term	In plain terms	In BC terms
Agent (Smart Agent)	A configured AI assistant users chat with to get work done in BC.	Everything that defines one assistant, gathered on one card.
AI model / LLM	The large language model that reads the message and writes the reply.	The "brain" the agent thinks with.
Model tier	A named quality/cost level of model (Fast, Smart, Expert, Premium).	Higher = better reasoning, more credits per message.
Credit multiplier	How many credits a tier burns relative to the baseline.	How much more a higher tier costs per message.
Token	The chunk of text (~¾ word) LLMs read and write; usage is measured in tokens.	The unit usage is counted in.
Credit	What you spend per message; tokens × tier multiplier.	A balance you draw down as agents work.
Prepaid / Subscription	The two billing modes: buy credits up front (Stripe) vs metered through Microsoft.	Prepay balance vs monthly metered invoice.
Instructions	The agent's always-on standing orders, read before every message. Also called the system prompt.	Default rules baked into a posting group or layout.
Capability	A coarse on/off gate: BC Data Access, Knowledge Base, Web Search, Image Analysis.	Top-level toggles on a permission set.
Tool	A specific action exposed to the agent — one API page, table, report, or web service it can call.	A single object permission line.
Tool calling / function calling	The agent deciding mid-conversation to run a tool, then using the result.	Drilling into a related table to get an answer.
Tool configuration	A reusable bundle of tools with read/write permissions.	A permission set, but for the AI.
Allow Create / Modify / Delete	Per-tool write permission: Not Allowed / Allowed / On Confirmation.	Permission set R/I/M/D flags, plus an approval option.
On Confirmation	The agent proposes a write; a human must approve before it commits.	An approval workflow step.
Usage description	A sentence telling the AI when to use a tool.	A field tooltip — guidance on when it applies.
Field rules / Tool Field Defaults	Hide, force, or pre-fill a tool's fields for the agent.	Field defaults on a template/config.
Parent-Child Field Links	Copy a key from a header tool to a line tool.	Document No. flowing from header to lines.
Autonomy level	How often the agent asks before acting: Low / Medium / High / Autonomous.	How many approval steps before posting.
Knowledge base	Documents you upload that the agent can read and quote.	Reference attachments the agent actually reads.
RAG (retrieval-augmented generation)	The behind-the-scenes way the app feeds relevant file passages to the model.	Auto-pulling the right reference doc into the answer.
Knowledge file	One uploaded document in the knowledge base.	One attached reference document.
Thread memory	A per-conversation scratchpad: goal, plan, verified data, decisions, errors.	A work-in-progress journal, cleared when the chat ends.
Session	One conversation thread between a user and an agent.	A single document/journal you're working in.
Output type (Chat / JSON)	Reply in natural language vs structured machine-readable data.	A printed report vs an API payload.
JSON schema	The structure a JSON-output agent must return.	A fixed report dataset definition.
Channel	Where users meet the agent: BC, Teams, or email.	Output method: screen / print / email / EDI.
Published / Publish	Making an agent available on a channel.	Publishing a web service or report.
Email channel	The agent monitors a mailbox on a schedule and auto-replies.	A recurring Job Queue task.
Suggested prompts	Starter questions shown at the top of a new chat.	Quick-launch tiles on a Role Center.
Department	A grouping label that organises agents in the chat sidebar.	Dimensions/grouping on a list.
Authorized users	The users allowed to use an agent when it isn't open to all.	The users in a permission/user group.
Test case	A saved conversation with an expected outcome.	A regression test scenario.
Test run	A replay of test cases, graded Pass / Partial / Fail.	Running the test suite after an upgrade.
LLM as judge	Using an AI model to grade whether a test passed.	An automated reviewer scoring the result.
Benchmark	Comparing models/configs against the same tasks.	A side-by-side performance comparison.
Backend	The QUALIA-hosted service that runs the models and provisions agents.	An external service endpoint BC calls.
Provision / Activate	Registering the agent on the backend so it can be used.	Releasing a document so it can be processed.
Tenant	One Business Central customer environment.	The environment/company boundary.
Share / Incoming share	Giving another tenant access to your agent / receiving one.	Distributing a configuration package.