Configure Gemini & Automate Salesforce Deals

Bring Your Own Model — Then Put It to Work Immediately

Eigent is model-agnostic. You can run it on the default cloud model, or you can connect any OpenAI-compatible API — including Google's Gemini — and use that instead. This walkthrough covers both sides: how to configure Gemini 3 Pro Preview in Eigent's settings, and what a real Salesforce automation task looks like when you run it through that model right away.

1Configure Gemini as a Custom Model

Open Settings → Models in Eigent and navigate to Custom Models. Select Gemini, then fill in:

• API Key: Your Google Gemini API key
• API Host: https://generativelanguage.googleapis.com/v1beta/openai/
• Model Type: gemini-3-pro-preview

Click Save. Eigent validates the model immediately and checks for function-calling support — a requirement for multi-step agent tasks. If validation succeeds, you'll see: "Validate success — The model has been verified to support function calling."

Set Gemini as your Default model, and every new task will run through it from this point on.

2The Task: Read Contact Data, Then Update a Deal

With Gemini configured, this prompt puts it to work immediately:

This is a two-part task that requires reading from one section of a Salesforce record (Contact Roles) and writing to another (Next Step). Eigent handles both in sequence.

3The Full Task Plan

Eigent generates a structured execution plan before touching Salesforce:

1. Locate the "salesforce.com - 200 Widgets" opportunity and move its stage from Needs Analysis to Proposal, then mark as current stage
2. Navigate to the Contact Roles section and retrieve the contact's name and phone number
3. Return to the main Opportunities view and edit the Next Step field with: book a meeting with [contact name] [phone number]
4. Save all changes

This plan is visible in the Eigent UI before any browser action is taken.

4Execution: Cross-Record Data Reading and Writing

The Search Agent opens Salesforce in the already-authenticated browser session. It navigates to the Opportunities tab, finds the "salesforce.com - 200 Widgets" record, and promotes the deal stage to Proposal.

The agent then moves into the Contact Roles subsection — a separate part of the same opportunity record — and extracts the contact name and phone number. With that data in hand, it returns to the top-level opportunity view and dynamically populates the Next Step field:

The record is saved. The entire workflow — stage update, data extraction, dynamic field population — runs without a single manual interaction.

5Why This Pattern Is Powerful

This workflow demonstrates a capability that most CRM automation tools struggle with: using live data from one part of a record to populate another. The contact's phone number isn't known at prompt time — Eigent retrieves it mid-task and uses it dynamically. That's a fundamentally different model from static field-mapping automation.

Combined with a custom model like Gemini, this workflow is also cost-efficient: you control the model, the API costs, and the rate limits, without being locked into a single provider.

6What to Try Next

7Tips for Better Results

• Validate before you run. The model validation step in Eigent's Settings catches incompatible APIs before you hit them mid-task. Always validate a new model configuration before using it in production workflows.

• Use precise record names. Including the exact deal name ("salesforce.com - 200 Widgets") eliminates ambiguity and prevents the agent from selecting the wrong opportunity in a large pipeline.

• Extend the pattern. Once you see how Eigent reads Contact Roles and writes to Next Step, the same approach applies to any cross-section data dependency in Salesforce: reading from a related list to populate a custom field, extracting product data to write a proposal summary, and more.