This article provides a comprehensive guide on building and using agents within the Vurvey platform.
What is an Agent?
Think of an agent as a smart intern within Vurvey. These AI-powered assistants can take on different roles, depending on your needs. An Assistant agent can handle specific tasks based on your instructions and provided data. A Persona (consumer-based segment representation) can take on the personality of a specific customer segment for you and respond as they would so you can chat with them and innovate together.
Unlike generic AI, Vurvey agents are context-aware, understand your workspace and objectives, and are capable of being grounded in your proprietary consumer responses from campaigns.
Key Features:
Contextual Understanding: Agents are trained to comprehend and work within your specific Vurvey workspace.
Task-Oriented: Designate specific tasks for your agents, such as summarizing information, writing content, or generating ideas.
Data-Driven: Provide agents with relevant data, such as market research or product specifications, to ensure tailored outputs.
Personalized Personas: Craft unique personas by defining an agent's personality, tone, and output structure.
Building Your First Agent:
1. Define the Purpose: Determine the specific role and tasks for your agent. For example, will it be a customer service representative, a market analyst, or a brainstorming assistant?
2. Craft the Persona: Outline your agent's personality, communication style, and desired output format. Consider factors like:
Name: Choose a name that reflects the persona and role.
Tone: Will the agent be friendly and conversational, formal and professional, or humorous and playful?
Output Structure: Should responses be concise bullet points, detailed paragraphs, or creative narratives?
3. Provide Instructions: Clearly outline the agent's tasks and responsibilities. Use specific language and examples to guide its behavior.
TIP - Use your AI Conversations feature in Vurvey to help you create your agent's background and instruction set. You can ask the AI to help you. Here is a sample prompt: "Train yourself as an expert at training and creating AI agent personas. Now create an agent background for Tommy, the master of Margaritaville who knows everything about mixed drinks and can help anyone make better margs."
4. Ground in Data: Upload relevant datasets to your workspace. These datasets will provide the agent with the knowledge base needed to generate informed outputs.
5. Test and Refine: Interact with your agent in draft mode, testing various prompts and refining its instructions based on the outputs.
6. Publish and Utilize: Once satisfied, publish your agent to make it accessible to your entire workspace.
Working with Datasets:
Datasets provide the foundational knowledge for your agents.
Creating Datasets:
Upload files in various formats (video, document, text, spreadsheet) directly to your workspace.
Provide a clear and descriptive name for each dataset.
Using Datasets:
When initiating a chat with an agent, specify the relevant dataset for it to reference.
Agents can analyze and extract insights from the provided datasets.
Tips for Success:
Start with a Clear Objective: Clearly define the agent's purpose and intended tasks.
Experiment with Creativity Levels: Adjust the creativity settings to control the diversity and imagination of the agent's responses.
Iterate and Refine: Continuously test, provide feedback, and adjust your agent's instructions to improve its performance.
Leverage the Power of AI: Don't hesitate to use the AI system to assist in generating agent backgrounds, instructions, and even image prompts.
FAQs:
Q: Are agents and datasets workspace-specific?
* A: Yes, any agent or dataset created within a workspace is only accessible to users with access to that workspace.
Q: Can I edit an agent after publishing it?
* A: Absolutely! You can edit, refine, and update your agents even after they've been published.
Q: How do I generate images within an agent chat?
* A: Use the command "/image" before your prompt to trigger image generation. For dedicated image generator agents, this command might not be necessary.
Q: Can I integrate web links directly into datasets?
* A: A web scraper tool is under development to enable this functionality soon.
Q: Can I export generated tables and charts?
* A: Currently, you can copy and paste them. Export options for Excel and CSV files are being explored.