To create and manage Context Sources, you need the Precursive AI Admin permission set and Salesforce Flow Builder access.

For configuring which compliance-categorised fields are permitted to flow through to the AI, see Configuring Compliance Category Filtering. 

What are Context Sources?

Context Sources define how to pull specific data from Salesforce for AI analysis. Each Context Source consists of two parts:

A Salesforce Flow that queries and returns the relevant data

• A Context Source record that registers the Flow with Precursive AI, describes the data, and lists the output variables

Once created, a Context Source can be reused across multiple Zen Agents and Rules. Builders select from available Context Sources when configuring Rules - they cannot create or modify Context Sources themselves.

How the Source Record Reaches the Flow

When a Zen Agent runs, it processes one source record at a time. The ID of that source record is automatically passed into the Flow as an input variable named recordId. From there, the Flow is free to:

• Query records that are directly related to the source record (e.g. tasks where WhatId = recordId)
• Query records that are unrelated to the source record (e.g. all open Opportunities across the account hierarchy)
• Ignore the source record entirely and return general org-wide data

The Flow always receives the source record ID, but what it does with it is up to you. This flexibility lets a single Context Source serve a wide range of use cases.

Salesforce is often the central hub for customer data. Even when you use other tools like Gong, Zendesk, or Gainsight, these applications typically sync data back into Salesforce. Context Sources let you tap into that data and make it available to the AI.

Examples of Context Sources:

• "Recent Tasks and Events" - activity history for any record
• "Related Contacts" - contact information linked to accounts
• "Support Cases" - cases synced from Zendesk or Service Cloud
• "Call Recordings" - transcripts and summaries synced from Gong
• "Customer Health Data" - scores and metrics synced from Gainsight

Important security considerations:

• Prompt injection risk: Prompt injection can occur where AI is provided with externally sourced or user-generated content (such as support cases or emails). Precursive mitigates this by controlling which data sources can be used as context, limiting exposure to untrusted inputs, and applying safeguards at the application layer. As an admin, carefully evaluate what types of user-generated content you include in Context Sources.
• External data sources: Where customers choose to extend AI context using external sources (e.g. MCP servers or APIs), these sit outside Salesforce's native access control model. As such, Precursive requires admin-controlled configuration and recommends only trusted, validated data sources, as equivalent RBAC controls may not apply to data retrieved from external systems.

Planning Your Context Source

Before building, consider:

• What data does the AI need? For example: related activities, contacts, products, open cases.
• Where does the data live in Salesforce? For example: Task, Event, Case, custom objects.
• Which Source Object will this work with? For example: Opportunity, Account, Project__c.
• Can this be reused? Design for reusability across multiple agents rather than building for one specific use case.

Review your data governance policies before creating Context Sources. You are in full control of what data is sent to the AI - nothing is shared automatically. Work with your compliance or legal team if you are unsure what data is appropriate to include.

Step 1: Create the Flow

Context Sources require a Salesforce Autolaunched Flow. The Flow must meet two requirements.

Input Variable: recordId

The Flow must accept an input variable named exactly recordId (case-sensitive).

Variable Name: recordId
Data Type: Text
Available for Input: Checked

Output Variables

The Flow must return data through one or more output variables.

Data Type: Record Collection
Available for Output: Checked
Variable Names: Use descriptive names such as recentTasks or openCases.
These names appear in the AI context and help the AI understand what the data represents.

Creating the Flow - Step by Step

1. Go to Setup > Flows and click New Flow
2. Select Autolaunched Flow (No Trigger) and click Create
3. Create the recordId input variable via Toolbox > New Resource > Variable
4. Add Get Records elements to query the data you need
5. Create output Record Collection variables with descriptive names
6. Add Assignment elements to assign the query results to your output variables
7. Connect all elements, save the Flow, and click Activate
8. Note the Flow API Name - you will need this when creating the Context Source record

Example: Recent Tasks and Events for an Opportunity

Get Records element 1 - Tasks: Object: Task
Filter: WhatId = {!recordId}
Sort: ActivityDate Descending

Get Records element 2 - Events: Object: Event
Filter: WhatId = {!recordId}
Sort: ActivityDateTime Descending

Output variables: recentTasks, recentEvents

Step 2: Create the Context Source Record

1. Go to the App Launcher and search for Context Sources
2. Click New
3. Fill in the fields:

4. Click Save

Step 3: Test the Context Source

There is no standalone preview surface for Context Sources today. To verify your Context Source is returning the data you expect, run it through a test Zen Agent and inspect the resulting context payload:

1. Create a test Zen Agent for the same Source Object
2. Add a Rule using your new Context Source
3. Use the Preview action to generate a test insight
4. Navigate to the Insight Result record and download the context file
5. Open the JSON to confirm the Flow returned the expected records and fields, and that the data matches what the AI is now being asked to analyse

Always test new Context Sources this way before they go into production agents.

Tips for Effective Context Sources

Design for reusability - prefix the name with the source object to make it easier to find and reuse across agents. For example, "Opportunity Recent Activities" or "Opportunity Account Events" rather than "Recent Activities" or "Sales Team Activities". Avoid hard-coding criteria that may not apply across all agents.

Keep data relevant - only gather data that will help the AI provide useful insights. Too much irrelevant data dilutes the output quality.

Use date filters - filter by recent date ranges such as the last 30 or 90 days rather than pulling all historical records. This keeps context focused and improves performance.

Document your Context Sources well - use the Context Description field to clearly describe what the data is, where it comes from, and any filters applied. Remember: this should describe the data, not how to use it. Per-agent usage instructions live on the Rule.

Troubleshooting

"Flow not found" error - verify the Flow API Name is spelled correctly (case-sensitive) and that the Flow is Activated and of type Autolaunched.

No data returned - test the Flow in Debug mode with a valid record ID. Verify filter conditions are not too restrictive and that related records exist for your test record.

Variable name errors - the input variable must be named exactly recordId. Output variable names must match what is listed in the Flow Output Variables field on the Context Source record.

Timeout or performance issues - reduce the amount of data being gathered by adding date filters, using record limits on Get Records elements, and filtering out low-value data. If a single Context Source is trying to gather too many different types of data, consider splitting it into multiple focused Context Sources - each assigned to its own Rule on the Zen Agent. Each Rule runs as a separate execution, which means each one has its own set of Salesforce processing limits to work within. Splitting large data-gathering logic across multiple Rules is the most reliable way to avoid hitting those limits on complex agents.