Using Filter Hints with AI Agents

Introduction

In this guide we will cover how to use Filter Hints in the Wabee AI platform! This feature allows you to provide additional context to AI Agents, enhancing their performance in specific scenarios by guiding their data processing.

---

What are Filter Hints?

Filter hints are optional parameters you can include in your API requests to suggest how the AI model should filter or focus on specific data. These hints help tailor the AI’s responses to better fit your needs.

Note: Filter hints are suggestions, not commands. The AI model may choose whether or not to apply them. Always test thoroughly before deploying to production.

---

How to Use Filter Hints

To utilize filter hints, include a data_filter_hints array in your API request. Each item in this array should be an object with the following properties:

Property	Description	Example
filter_key	The field name to filter on	CompanyName
filter_type	The type of filter operation (e.g., "equals", "gt", "lt", "gte", "lte", "like")	equals
filter_value	The value to filter by	Acme Co.
tool_name	The name of the tool or data source this filter applies to	bubble_data

Example API Call

Here's a sample code of how you can implement the API call using JavaScript with the fetch API:

const apiUrl = 'https://my-agent-id.wabee.ai/core/v1/chain/';
const payload = {
  messages: [
    {
      content: "What are the top 10 recent sales?",
      role: "user"
    }
  ],
  data_filter_hints: [
    {
      filter_key: "CompanyName",
      filter_type: "equals",
      filter_value: "Acme Co.",
      tool_name: "bubble_data"
    }
  ]
};

fetch(apiUrl, {
  method: 'POST',
  headers: {
    'Content-Type': 'application/json',
    'x-wabee-access': 'YOUR_API_KEY'
  },
  body: JSON.stringify(payload)
})
.then(response => response.json())
.then(data => {
  console.log('AI Agent Response:', data);
})
.catch(error => {
  console.error('Error:', error);
});

Explanation:

1. Endpoint: The API endpoint is set for example purposes only. Replace https://my-agent-id.wabee.ai/core/v1/chain/ with your agent endpoint from the agent settings page.
2. Headers:
3. Content-Type: Indicates that the request body is in JSON format.
4. x-wabee-access: Your API key for authentication.
5. Payload: Contains the user message and the data_filter_hints.
6. Fetch API: Sends a POST request with the specified headers and payload.
7. Response Handling: Logs the AI Agent's response or any errors to the console.

---

Use Cases

1. User-Specific Data
2. Scenario: Integrate AI Agents into your product's chat interface.
3. Implementation: Use filter hints to provide context based on the logged-in user's permissions or preferences.
4. Time-Based Filtering
5. Scenario: Focus the AI on recent data.
6. Implementation: Apply a filter hint with a date field to prioritize newer information.
7. Department or Category Filtering
8. Scenario: Large organizations with multiple departments.
9. Implementation: Use filter hints to target data relevant to specific departments or categories.

---

Best Practices

1. Test Thoroughly
2. Ensure your implementation works across various scenarios.
3. Don't Overly Rely on Filters
4. Prepare your application to handle cases where the AI might not apply the filters as expected.
5. Use Meaningful Filters
6. Select filter hints that are relevant and beneficial for the AI’s tasks.
7. Monitor Performance
8. Track how filter hints impact your Agent's performance and make adjustments as needed. When filters are used by the agent in tools, the ability of the tool to apply the filter efficiently is dependent on the tool's capabilities.