Capture from AI action

The Capture from AI action is an advanced feature that allows you to enrich and structure data captured in your agent by passing input text through an AI model (such as OpenAI, Claude, Ollama, vLLM, or Google Cloud Vertex AI / Gemini). The AI service processes complex, unstructured text and extracts or transforms it into defined, structured data fields.

Important Note: Accuracy of AI-generated results is not guaranteed. It is recommended to test and use this feature carefully. AI models excel at extracting facts from free-form text, generating summaries, performing sentiment analysis, translating languages, standardizing dates, or parsing complex unstructured descriptions. *

Steps for Setting Up Your Account for AI Integration

Before using the Capture from AI action in Agent Builder, you must configure your credentials or server endpoints in the Mozenda Web Console.

1. Obtain Credentials or Server Endpoints:

a. OpenAI: Set up an account and generate an API key at openai.com.
b. Claude: Set up an Anthropic account, generate an API key, and optionally obtain your Workspace ID for organizational access.
c. Ollama: Deploy or locate your hosted Ollama server endpoint URL.
d. vLLM: Deploy or locate your hosted vLLM server endpoint URL and generate an Auth Token.
e. Vertex AI (Google Cloud Platform / Gemini): Create a GCP project, specify the region, and generate a Service Account JSON key with access to Vertex AI services.

2. Configure Web Console Settings:

a. Close the Agent Builder if it is currently open.
b. Log in to the Mozenda Web Console.
c. Navigate to the Account Settings Summary page.
d. Select AI Configurations from the left navigation page menu.

3. Enter AI Credentials:

a. In the Manage AI Configurations modal, fill in the relevant credentials for your provider(s):

i. OpenAI: Insert your Api Key.
ii. Claude: Enter your Api Key and optional Workspace Id.
iii. Ollama: Enter your Server URL (e.g., https://llm.mozenda.com/api/chat).
iiii. vLLM: Enter your Server URL (e.g., https://vllm.mozenda.mobi/v1/chat/completions) and Auth Token.
iiiii. Google Cloud Platform (Vertex AI): Enter your Project ID, Region (e.g., us-central1), and paste your Service Account JSON payload.

b. Click Save.

Setting Up a Capture from AI Action in Agent Builder

1. Open the Agent Builder and open an existing agent or start a new agent.

2. Create a Capture Text action for the source text you want to evaluate/pass to the AI model.

3. Add a Capture from AI action to your action list.

Action Parameters & Configuration Options

When configuring the Capture from AI action modal, complete the following fields:

• Input: The text field or variable containing the raw data to be sent to the AI model. Click Select to choose a captured field or item.

• Optimize input text before uploading (Checkbox): When checked, Mozenda automatically optimizes the input text by stripping unnecessary words, formatting, or tokens. This reduces overall token size, lowers API costs, and speeds up response times.

• Provider: Select your preferred AI provider backend from the dropdown menu:

  • OpenAI
  • Claude
  • Ollama (supports custom or self-hosted open-source models)
  • vLLM (supports custom or self-hosted open-source models via vLLM endpoint)
  • Vertex AI (Google Cloud / Gemini models)
    

• Model: Select or enter the specific AI model to process the request (e.g., claude-haiku-4-5-20251001, gpt-4o, gemini-1.5-flash, or custom Ollama/vLLM model names such as Qwen/Qwen3-0.6B). Selecting a provider populates a default model, which can be modified as needed.

• Prompt: Instructions directing the AI on how to interpret or transform the input text. The default system prompt is generally sufficient for structured extraction into JSON.

• Output Fields: Define the key-value target fields to save from the AI response:

  • Name: The field name where extracted data will be saved (e.g., Age, Size, Sentiment).
  • Description: A simple description explaining what data the AI should return for that specific field (e.g., The age of the person as an integer or Extract size from text).
  • Click Add to add additional output fields.

4. Click Create Action.
5. Test the Agent: Test the agent step to execute the request and receive data from the selected AI service.

Refining Captured AI Data

After executing the Capture from AI action during testing, you can use Refine Captured Text to review and adjust how the returned structured data maps to your agent's fields.