Terms as they are used in AI Engine and in these articles.
Action key. The key that links a prompt template to an action. Using a built-in key (summarize, nextSteps, draftEmail, companyNews, webCheck) replaces the built-in prompt for that action. See Prompt templates.
Actions bar / Actions menu. The bar below the record banner in the chat, with the built-in actions and the ◈ Actions menu of custom prompts. See Actions menu.
Activity Digest. The answer to a general "what happened this week?" question: counts and recent records across the objects enabled for the digest.
Activity Monitor. The admin dashboard of AI usage: calls, tokens, estimated cost, users, features and errors.
AI Activity Log. The object that stores one record per AI action with usage metadata, plus a short error message when a call fails. It does not store full prompts or answers.
AI Score. A 0–100 score with a label and reasoning, stored in AI_Score__c, AI_Score_Label__c, AI_Score_Reason__c and AI_Score_Date__c on Lead and Opportunity.
Agent Objects. The admin setting that decides which objects and fields the chat can use. Chat questions can only query enabled objects, and the AI is told only about their enabled fields. Some features have their own field settings, such as Scoring Context Fields for AI Scoring. See Agent Objects.
Allow PII. Org Profile setting. Off (the default) means email, phone and address field values, fields you mark as PII, and recognisable patterns such as email addresses and IBANs in free text are replaced with placeholders before a request is sent. Other personal details written in free text, such as names, are not masked. See PII masking.
Bring your own key. You hold the AI provider account and the API key. AI Engine calls the provider through a Named Credential in your org. The key stays in your org's credential settings and is not sent to Cloud Connected.
Cached tokens. Input tokens the provider serves from its prompt cache, usually at a lower price. Shown as Tokens Cache Read and Tokens Cache Write in the log.
Child objects. Related objects whose records are added to a record's context, configured per object in Agent Objects. Meet Prep can add further related objects, set in the Org Profile under Meeting Prep — Additional Objects.
Custom PII fields. An Org Profile list of field API names whose values are masked when Allow PII is off.
Data source. A saved query attached to a prompt template. Its rows are added to the prompt where the {data:name} placeholder appears. See Prompt data sources.
Estimated cost. AI Engine's estimate of provider cost per action, based on a built-in price list. Your provider's invoice is authoritative. See Token usage and cost.
External Credential. The Salesforce Setup item that stores your provider API key. It is used through a Named Credential.
Fast model. A cheaper model from the same provider (Fast model (internal steps) in the Org Profile), used for web-search steps and Help mode. If it is left blank, web-search steps with OpenAI or Anthropic use a built-in cheaper model, and Help mode uses the main model.
Global prompt. A prompt template with Global scope. It appears in the Actions menu when the chat is opened without a record, and runs without record context.
Help mode. Chat mode switched on with the ? icon. Questions are answered from the built-in help instead of your data.
IN CONTEXT / DETACHED. The record badge in the chat. IN CONTEXT sends the current record with your question; DETACHED leaves it out.
Main model. The AI Model in the Org Profile, used for chat, record actions, scoring and other main work.
Max Context Chars. The Max Context Chars (Data Sources) field of the AI Engine Profile record, which caps the total text data sources can add to one prompt (default 40,000 characters). It is not shown in the admin panel.
Max Output Tokens. Org Profile setting for the length of AI output in record actions. Most record actions use at least 3,500 tokens even if it is set lower. See Limits.
Named Credential. The Salesforce Setup item that holds the provider URL and points to the External Credential. You type its name in the Org Profile.
Org Profile. The org-wide AI settings: provider, models, Named Credential, playbook, PII settings and branding.
Placeholder (token). The value that replaces masked personal data in a request, such as EMAIL_1 or PHONE_2. Real values are put back into the answer.
Playbook. The Org Profile texts (Company Description, Lead Process, Email Guidelines, Qualification Questions, Compliance Guardrails) that are included in chat and record-action requests to shape tone and content.
Principal access. Access to an External Credential's principal, granted through a permission set. Users need it to make AI requests.
Record prompt. A prompt template with Record scope. It runs on a record and receives its context, optionally only on selected objects.
Scoring Context Fields. The fields the AI sees when it scores a record. See Scoring context fields.
Self-repair. When a query written by the AI fails, AI Engine sends the error back to the AI once for a corrected query and runs it again.
Show SOQL in Chat. A troubleshooting option that shows the query behind each chat answer. It is saved in the browser where it is switched on.
Token. The unit AI providers measure and bill usage in. As a rough guide, a token is a few characters of English text; other languages usually need more tokens for the same content.
Web search. The provider's web search tool, used by Web Check, Company News, research questions, Meet Prep web search, Verify Company On The Web and Magic Snap research.