Setup#
Does AI Engine work in Salesforce Classic?#
No. All AI Engine components are Lightning components and need Lightning Experience.
Do we need our own AI provider account?#
Yes. You use your own OpenAI, Anthropic or Azure OpenAI account and key, and the provider bills you directly for usage. Cloud Connected does not supply or hold keys. See Connect your AI provider.
Can we switch provider later?#
Yes. Create credentials for the new provider and change AI Provider, AI Model and Named Credential in the Org Profile. No reinstallation is needed, and your configuration stays. AI Engine does not switch provider by itself, but keeping credentials for a second provider ready lets you switch quickly during an outage.
Why does the setup ask for "CloudConnected" in the Named Credential?#
AI Engine runs in the CloudConnected namespace. Salesforce lets a managed package use a Named Credential you created only if its namespace is listed in Allowed Namespaces for Callouts (the Managed Package Callout Allowlist section of the Named Credential).
What happens when the trial ends?#
AI actions are blocked and the chat shows that the trial has expired. Your configuration, templates, scores and data stay as they are. Enter a license key from Cloud Connected in Ai Agent Admin Setup → License to continue. For pricing, contact Cloud Connected. See License and trial.
Data and permissions#
Can a user see records through AI Engine that they cannot open in Salesforce?#
The chat's queries run with the user's own sharing, object permissions and field-level security, and only on objects the admin enabled in Agent Objects (plus tasks, events and users), so they return only records the user can already access. Text that AI Engine writes into a field, such as a score reason, is visible to anyone who can read that field, like any other field value. See Permissions and sharing.
Does AI Engine change our data on its own?#
AI Engine creates or changes your business records only when a user clicks: creating a task, sending an email, downloading, saving or emailing a PDF (each saves the PDF to Files on the record), saving a Magic Snap record, generating a score on the AI Score card. The exception is AI Scoring: when an admin turns on automatic or scheduled scoring, AI Engine writes the score fields on Leads and Opportunities. AI Engine also keeps its own records, such as the AI Activity Log, as it runs.
Are the requests sent to the AI and its answers stored?#
No. AI Engine does not save the text it sends to the AI provider or the answers that come back. Your prompt templates (the actions in the Actions menu) are configuration and stay saved in your org. The AI Activity Log keeps usage metadata (feature, user, object, model, tokens, duration, status), plus a short error message when a call fails. That error message comes from the provider and can quote part of the request. Cloud Connected does not receive prompts or answers: requests go from your org straight to your AI provider. The provider's own retention of API requests (for example for abuse monitoring) is set by your agreement with that provider. See How AI Engine handles your data.
Where is our data processed?#
By the AI provider you chose, under your account with them. With Azure OpenAI, that is your own Azure resource and region, except for the features that search the web, which use OpenAI. See Choosing a provider for compliance.
Using AI Engine#
Why does the same question give slightly different answers?#
AI models do not always phrase answers the same way. The underlying data query is usually the same; turn on Show SOQL in Chat in the Org Profile to compare the queries. That switch applies only in the browser where you turn it on, not to other users. A lower Temperature in the Org Profile makes answers more consistent on models that support it.
Why is my answer empty?#
For record actions such as Next Steps, Draft Email or Meet Prep, the output limit is usually too small for the answer, especially in languages other than English. These actions use Max Output Tokens from the Org Profile or 3,500 tokens, whichever is higher, so set it above 3,500. Chat answers are capped at 1,000 tokens regardless of the setting; ask a narrower question instead.
Why does the chat only show 200 records?#
Chat results show the first 200 matching records. The table says how many are shown, and Show all loads up to 2,000. CSV and Excel exports include up to 10,000 rows; for larger result sets use a Salesforce report.
Which languages are supported?#
The chat answers in the language of your question. Record actions and Meet Prep follow the language selector in the chat (EN, PL, DE, FR, NL): English, Polish, German, French or Dutch. The admin interface is in English. See Languages.
Does the chat remember what I asked before?#
Yes, the last few messages. A follow-up like "and only from Germany" refines the previous question. Click ✕ Clear to start over.
Can many users use AI Engine at the same time?#
Yes, within Salesforce limits. Salesforce limits how many long-running requests an org can run at the same moment. With many users starting AI actions at once, some may see "Concurrent requests limit exceeded" and need to retry after a few seconds. Running scheduled scoring outside business hours helps. See Limits.
Cost#
How do we see what AI Engine costs us in provider fees?#
The Activity Monitor shows calls, tokens and an estimated cost per month, feature and user. The estimate is based on the model used for each call. Your provider's invoice is the authoritative figure. A model name that is not in AI Engine's built-in price list, such as a custom Azure deployment name, is estimated with a default rate, so its cost figure is only approximate; its tokens are counted exactly as reported by the provider. See Token usage and cost.
What makes usage more expensive?#
Many fields per object in Agent Objects, Meet Prep with web search (much larger than a normal action), and automatic scoring on large imports. Keep the field lists focused and use the scoring filter (Filter — Which Records to Auto-Score in AI Scoring).