The Org Profile holds the AI settings that apply to every user in the org: which provider and model AI Engine calls, how long answers may be, whether personal data is masked, and the company playbook that shapes most answers. This article is a reference for each setting.
Before you start#
- Create the External Credential and Named Credential for your AI provider in Salesforce Setup first. See Connect your AI provider.
- You need the
AI_Engine_Adminpermission set.
Open or create the profile#
- Open the Ai Agent Admin Setup tab and select Org Profile.
- If a profile card is listed, click it to edit it. If the tab says No profile yet, click + New Profile.
- Change the settings, then click ✓ Save at the top of the form. Cancel leaves without saving.
AI Engine works with one Org Profile. The profile card shows the main model and an ACTIVE badge.
Always edit the existing profile by clicking its card. If a profile already exists, saving from + New Profile does not add a second profile. It overwrites the existing one with the values of the new, mostly empty form, including the playbook, branding, custom PII fields and Meeting Prep objects.
AI settings#
| Setting | What it does | Default for a new profile |
|---|---|---|
| AI Provider | OpenAI (Responses API), Azure OpenAI or Anthropic Claude. Changing the provider resets AI Model and Named Credential to that provider's defaults, so check both afterwards. | OpenAI (Responses API) |
| AI Model | The main model, used for chat answers and the queries behind them, record actions, Meet Prep, PDF text and scoring. For OpenAI and Anthropic, pick from the list. For Azure OpenAI, type your deployment name. | The provider's recommended model. For Azure OpenAI the field is prefilled with gpt-4o; replace it with your deployment name. |
| Fast model (internal steps) | An optional cheaper model from the same provider, used for web searches and the in-app help assistant. A model from another provider is not saved. For Azure OpenAI, this is an optional second deployment name. See Choose the main and fast models. | Provider default (for Azure OpenAI: blank, which means the main deployment) |
| Named Credential | The name of the Named Credential you created in Setup, spelled exactly as in Setup. | OpenAI, Anthropic or AzureOpenAI, depending on the provider |
| Max Output Tokens | Raises the output limit of record actions, Meet Prep, PDF templates and Analyze Email above their built-in minimums. See below. | 1000 |
| Temperature | How varied the wording is. Lower values give more consistent answers. | 0.5 |
| Allow PII | Off: AI Engine masks personal data before it is sent to the AI provider. This covers email, phone and address fields, the fields you flag as PII in Agent Objects, the custom PII fields below and some recognisable patterns in free text. On: values are sent as they are. See PII masking. | Off |
| Show SOQL in Chat | Shows the query behind each chat answer. Useful when you test the agent. | Off |
| Modern Interface | Switches the chat to the modern theme. See Branding. | Off |
| Meet Prep: include web search | Adds a web search for recent company news to every Meet Prep brief. | Off |
Max Output Tokens#
Record actions such as Next Steps, Draft Email and your own prompt templates, Meet Prep briefs and the AI text in PDF templates always allow at least 3,500 output tokens, and Analyze Email at least 2,000. Max Output Tokens takes effect only when you set it higher than those values; it then raises the limit for those features. Chat answers are capped at 1,000 tokens, and AI scoring, Magic Snap and the help assistant have their own fixed limits. This setting does not change them.
If long briefs or results from these features stop mid-sentence, or structured results such as cards come back empty, set a value above 3,500. Answers in languages such as Polish, Dutch or French use more tokens than the same answer in English.
Temperature#
Keep it between 0 and 1. For sales work, 0.3 to 0.7 is a good range. Some newer models do not accept a temperature setting. For those models, AI Engine leaves it out of the request.
Show SOQL in Chat#
This switch is stored in the browser of the admin who changes it, not in the org. It shows queries only to that admin, in that browser. Use it for testing. It does not turn on query display for your users.
Meet Prep web search#
A web search adds roughly 30,000 input tokens to each brief. That is why it is off by default. Switch it on only if your reps want recent news in every brief.
Custom PII fields#
Custom PII fields (masked when Allow PII is off) takes a comma-separated list of field API names, for example Tax_ID__c, National_ID__c. These fields are masked in addition to the fields you flag in Agent Objects. For how masking works, see PII masking.
Playbook sections#
The five text areas Company Description, Lead Process, Email Guidelines, Qualification Questions and Compliance Guardrails together form the playbook. What to write there is covered in Write the playbook.
Branding#
App Name and Logo URL in the Branding card change the header of the Sales Agent chat for all users. See Branding.
Meeting Prep: additional objects#
The 📅 Meeting Prep — Additional Objects card at the bottom of the profile selects which related records Meet Prep reads. It is saved with ✓ Save, together with the rest of the profile. See Child objects and Meeting Prep.
Troubleshooting#
Every AI call fails straight after setup. Check three things in Salesforce Setup:
- The Named Credential value in the Org Profile matches the Named Credential's name in Setup exactly.
CloudConnectedis listed under Allowed Namespaces for Callouts (the Managed Package Callout Allowlist section) on the Named Credential.- Every AI Engine user, admins included, has access to the External Credential's principal through a permission set of your own (Setup → Permission Sets → that permission set → External Credential Principal Access), and also holds AI Engine User.
Record actions, Meet Prep briefs or PDF text stop mid-sentence. Set Max Output Tokens above 3,500.
The playbook or branding is suddenly empty. The profile was probably saved from + New Profile, which overwrites the existing profile. Enter the values again on the existing profile card.