Weflow runs its AI on third-party foundation models, primarily Google Gemini, and never uses your data to train them. The AI reads only the Salesforce data a user may already see, and admins decide which teams, objects and assistants can reach it.
Which AI models Weflow uses with your data
Weflow uses established foundation models through their APIs and hosts no model of its own. Your Data Processing Agreement (DPA) states that customer data is never used for model training, and Weflow adds an explicit no-training provision to your DPA on request.
No data is retained or stored with Google Gemini, and no model is ever trained with customer data. Weflow uses Gemini on Google Cloud.
What Weflow AI does with your data
Feature | What it reads | What it produces |
AI Field Updates | Call transcripts | Values in Salesforce fields |
AI follow-up emails | Call transcripts | Drafts you review before sending |
AI Playbooks | The whole CRM record | Qualification status per record |
Ask Weflow AI | Salesforce, emails, meetings, public web | Answers to your questions |
Ask Weflow AI does not reach your other knowledge systems, such as process documentation or a ticketing system.
How Weflow AI respects your Salesforce permissions
Ask Weflow AI signs in with each user's own Salesforce token and inherits their Salesforce permissions, so it returns only data the user can already access. It then applies its own guardrails on top, and can withhold information a user could reach directly in Salesforce.
How admins control Weflow AI access to your data
Ask Weflow AI access: switch it on for Everyone in the organization or for Specific teams only.
Context and Sources: choose which Salesforce objects, standard and custom, Weflow AI may read.
MCP connector: Enable MCP Connector lets your team connect an AI assistant, such as Claude, to search recordings, pull forecasts and query pipeline data. Allow transcript access decides whether connected assistants read full call transcripts or only AI summaries. The page lists every connected user with their email, when they connected and when they last used it.
Read-only mode: blocks every field update from Weflow into Salesforce, AI Field Updates included, while capture and recording keep working.
Switching AI features off: an admin can switch off the MCP Connector, Ask Weflow AI, AI Playbooks, AI Coaching Scorecards, AI meeting summaries, AI Field Updates and Agents.
Learn more in How to Set Up Ask AI for Your Organization and Weflow MCP Server Overview.
How Weflow protects AI processing of your data
Prompt injection: Weflow separates system instructions from your data, sanitises inputs and lets the model return only structured results that cannot run actions.
Tested: Weflow's annual third-party penetration test covers its AI endpoints.
No source label: a value Weflow writes to a Salesforce field is not marked with the source it came from; the field update view shows what changed.

