AI Agent

viaSocket AI Agents built-in tool move beyond traditional workflows by leveraging natural language instructions and AI decision-making. Instead of rigid branches, you define:

• A trigger

• AI-driven instructions (written in plain English)

• Connected data sources

• Automated actions across ~1,500 integrations

This makes the agent adaptive, scalable, and less reliant on hardcoded logic.

Why Use an AI Agent?

Here’s what you’ll love:

• Smart & Adaptive: Understands the intent behind triggers and adapts its response.

• Data-Aware: Can access and use your uploaded files, app data, and CRM info.

• Multi-App Actions: Can create replies, update records, send messages, and more.

• No Code: You don’t need to write scripts—just describe the behavior.

Step-by-Step Breakdown :

1. Starting with Templates or Scratch

• Users can either use pre-built templates (like a "Support Email Agent") or build from scratch.

• Triggers are selected from existing apps (e.g., Gmail, Outlook, HubSpot).

2. Insert Actions

Actions are what the agent does after understanding the situation.

Examples:

• Draft an email

• Send a Slack DM

• Update a Google Sheet

• Create a support ticket

3. Configure the viaSocket AI Agent

1. Select LLM Service

From the LLM Service dropdown:

• Choose the AI provider you want to use (e.g., OpenAI, Google Gemini, Anthropic Claude).

• This determines how your API call is routed.
  

  
  

2. Enter API Key

• Paste your API Key securely into the field.

• This key authenticates your request to the selected LLM.

💡 Tip: Use environment variables or secrets manager for sensitive keys.

3. Define Prompt

This is the core instruction to guide the AI’s behavior. Think of it like this:

Example instruction:

When I receive a new email, check if it’s a customer query. If it is, look up the answer in my “FAQ doc.” If you find an answer, draft a polite reply. Also, send a Slack message to @support with the draft.

More Examples:

• Summarize the customer’s issue.

• Classify the query based on intent.

• Generate a response using FAQ data.

4. Optional: Thread ID

Use this field if:

• You're maintaining conversational context over multiple calls.

• You want the model to remember prior exchanges using thread/session IDs.

If not needed, you can leave it blank.

5. Optional: Query

You can pass a specific user input or question into this field—ideal when you're pulling dynamic values from previous steps.

Example:

• What is the refund policy?

6. Add Variables (Dynamic Inputs)

These let you pass structured data or custom values into your prompt using variable references.

To add:

1. In the first field, enter the variable name (e.g., topic, user_email, customer_message)

2. In the second field, enter the value (static or dynamic from a previous step)

3. Click Add

You can then use these inside your prompt like:

plaintext

Summarize the following message: {{customer_message}}

7. Test the Agent

Before saving, click Test to:

• Send a test call to your selected LLM

• View the live RESPONSE from the model

• Debug any prompt or variable issues

8. Create & Save

If the output looks correct:

• Click the Create button to finalize the AI Agent block in your workflow.

Example Use Cases