This is the shortest path from a fresh installation to a working AI Engine. It is written for the Salesforce admin doing the setup. Each step links to a detailed article.
Before you start#
- AI Engine is installed. See Install the package.
- You have an API key from OpenAI, Anthropic or Azure OpenAI.
Step 1: Assign permission sets#
In Setup → Permission Sets, assign AI Engine Admin and AI Engine User to yourself, and AI Engine User to the people who will use AI Engine. Admins need both. See Assign permission sets.
Step 2: Connect your AI provider#
Create an External Credential that holds your API key and a Named Credential that points to the provider's URL. Add CloudConnected under Allowed Namespaces for Callouts on the Named Credential, and grant access to the credential's principal through a permission set of your own, assigned to everyone who uses AI Engine. See Connect your AI provider.
Step 3: Open the admin tab#
Open the Ai Agent Admin Setup tab (find it in the App Launcher and pin it to your app's navigation). The License tab inside it shows the trial status. The 5-day free trial starts the first time someone uses an AI feature, for example a chat question, not when you open this tab.
Step 4: Configure the Org Profile#
In Ai Agent Admin Setup → Org Profile, click + New Profile and set:
- AI Provider: OpenAI (Responses API), Azure OpenAI or Anthropic Claude.
- AI Model: the main model. For Azure OpenAI, type your deployment name.
- Named Credential: the Name (API name) of the Named Credential from step 2, not its label. It must match exactly, including upper and lower case.
- Max Output Tokens: new profiles start at 1000, which is fine to begin with. Most record actions use at least 3,500 output tokens whatever you enter here, so raise this value only if long answers from record actions are cut off.
- The playbook sections (Company Description, Lead Process, Email Guidelines, Qualification Questions, Compliance Guardrails): a few sentences each about your company, products, sales process and tone. AI Engine adds this text to chat requests and to the built-in record actions.
Leave Allow PII off unless you have decided otherwise. See Org Profile and PII masking.
Click Save.
Step 5: Choose what the agent can see#
In Agent Objects, select the objects your team works with, for example Account, Contact, Lead and Opportunity, plus your two or three most important custom objects. For each one, tick the fields the agent may read, click Mask as PII next to each ticked field that contains personal data, switch on Active for Agent and click Save. The chat can query only the objects you enable here (plus tasks, events and users), and it works from the fields you tick. Start with a focused list; you can extend it later. See Agent Objects.
Step 6: Add the components to pages#
In Lightning App Builder, add salesagent to the record pages your team uses (and to the utility bar if you want the chat available everywhere), and aiScoreCard to Lead and Opportunity record pages. Save and activate each page. See Add components to pages.
Step 7: Test as a real user#
Log in as (or use Login As for) a user who has only AI Engine User, plus access to the credential's principal, and run a few questions and actions. See Run a first test.
Optional next steps#
- AI Scoring for Leads and Opportunities.
- Prompt templates for your team's recurring tasks.
- PDF templates and PDF settings.
- Roll out to users.