Creating a Gemini Agent
Estimated Time: 2 minutes
A Gemini resource connects your agent to Google Gemini models using your own Gemini API key. Predictable Dialogs supports the latest Gemini models exposed in the model selector.
Before You Start
You need a Gemini API key.
- Open the Google AI Studio API keys page.
- Create or copy an API key.
- Make sure the Google project has billing and model access for the model you want to use.
Create From The Agents Page
Step 1: Create New Agent
- Sign in to Predictable Dialogs.
- Click + Create agent.
- Give a name to your agent.
Step 2: Configure Gemini Settings
- Click ADD AI.
- Select Gemini as the provider.
- Select an existing Gemini API key or enter a new API key.
- Select the Gemini model you want to use. The latest supported Gemini models are listed in the model dropdown.
- Click Next to validate the key and model.
Step 3: Add The Instructions
- Choose what your agent is for and adjust the personality and purpose.
- Click Next.
- Review or edit the agent behaviour instructions.
- Click Save to create the resource and connect it to your agent.
Understand The Instruction Hierarchy

The Agent instructions section shown at the top is closed in the screenshot. It contains the provider and model details, along with the agent personality and purpose.
The Skill Instructions card contains the agent behaviour. It also contains Knowledge & actions and the Add button used to add documents, tools, and actions the agent can use. This separation is intentional: the agent personality stays separate from the skill, which defines what the agent knows how to do.
Next Steps
- Add knowledge and actions: Use the Tools overview to add searchable documents, API tools, app connections, or website payload actions to the agent skill.
- Choose Channel: Embed your agent on your website or connect to WhatsApp.
- Test Your Agent: Send test messages to verify everything works correctly.