Relevance AI Review: Building AI Agents and Multi-Agent Systems

Relevance AI is a no-code platform for building AI agents and multi-agent systems that has been gaining attention in the AI development space. This review provides an in-depth look at the platform's features, capabilities, limitations, and use cases based on extensive testing.

Overview of Relevance AI

Relevance AI differentiates itself from other agent-building tools by making it exceptionally easy to create multi-agent teams without writing code. What might take weeks to build using a traditional agent framework can often be prototyped within hours using Relevance AI and deployed within a day or two.

The platform allows users to quickly build AI agents with capabilities like:

• Natural language interactions
• Integration with external tools and data sources
• Multi-step workflows
• Collaboration between multiple specialized agents

While powerful and user-friendly, Relevance AI does have some limitations around customization and deployment options that are important to consider.

Key Features and Capabilities

Core Instructions

The core instructions are where you define the overall purpose and behavior of an agent. This can range from very detailed instructions to more concise directives, depending on how much structure is provided elsewhere in the agent configuration.

The core instructions tell the agent what it's trying to accomplish at a high level. For complex agents that need to handle a variety of user inputs and scenarios, more detailed instructions are often beneficial.

Flow Builder

The flow builder allows you to create a structured sequence of steps for your agent to follow. This is particularly useful for agents that need to gather specific pieces of information or follow a defined process.

For example, you might use the flow builder to ensure an agent captures certain data points from a user before triggering a database search. The flow builder helps guide the agent's conversation in a more controlled manner.

Tools

Tools give agents the ability to take actions outside of Relevance AI. Some examples include:

• Running Google searches
• Interacting with databases
• Scraping websites
• Calling external APIs

Relevance AI provides a variety of pre-built tools that can be easily added to agents. Users can also create custom tools to extend agent capabilities.

When adding tools, you can specify whether the agent needs to ask for permission before using them or if they can be auto-approved. This provides control over potentially sensitive or resource-intensive operations.

Sub-Agents

One of Relevance AI's standout features is how easily it allows users to build multi-agent systems. You can simply select other agents to make available as sub-agents, allowing them to collaborate on complex tasks.

Each sub-agent has its own core instructions, flow, and tools, allowing for specialized capabilities within a larger system. This makes it possible to create sophisticated AI systems composed of multiple specialized agents working together.

Integrations

Relevance AI offers integrations with various external services and APIs. Some notable integration options include:

• Text-to-image, text-to-audio, and text-to-video services
• Data processing tools
• CRM platforms like HubSpot and Salesforce
• Google Calendar
• Zapier for connecting to additional services

While the range of native integrations is somewhat limited compared to some other platforms, the Zapier integration and custom API capabilities help fill gaps for many use cases.

Deployment Options

Once an agent is built, Relevance AI provides a public URL and iframe embed code for sharing. This allows for quick deployment, but customization options are limited:

• The agent interface appears as a chat panel
• Look and feel cannot be extensively customized
• No native options for creating fully custom UIs

For internal tools and prototypes, the provided deployment options are often sufficient. However, they may be limiting for customer-facing products that require a more polished, branded experience.

Language Models

Relevance AI supports a range of language models from OpenAI and Anthropic, including:

• GPT-3.5 and GPT-4 models
• Claude and Claude Instant

While not the most extensive selection of models, it covers many of the most popular and capable options currently available.

Building Agents with Relevance AI

To illustrate how Relevance AI works in practice, let's walk through the process of building two example agents: a travel planning assistant and a financial advisor.

Travel Planning Agent

This multi-agent system helps users plan trips, with a focus on family travel. Here's how we set it up:

1. Create a new agent named "Travel Agent AI"
2. Set core instructions defining the agent's goal and key steps (understand destination/hotel, find flights, handle special requests, etc.)
3. Add tools:
   • Google Search
   • Website scraper for hotel information
   • Custom "Travel Destination Finder for Parents" tool
4. Build a flow to guide the conversation:
   • Ask for initial travel preferences
   • Use the destination finder tool to suggest options
   • If an option is chosen, use a flight-finding sub-agent
   • Scrape hotel website for contact info
   • Offer to draft an email to the hotel
5. Add the flight-finding sub-agent

Testing the agent shows it successfully gathering travel preferences, suggesting destinations, finding flights, and offering to contact hotels - demonstrating the power of combining multiple tools and sub-agents.

Financial Advisor Agent

This agent aims to provide cash flow modeling and financial forecasting. It illustrates some of the challenges in building more complex agents:

1. Create a new agent with detailed core instructions covering:
   • Understanding the user's current financial situation
   • Analyzing cash inflows (income, investments, etc.)
   • Analyzing cash outflows (expenses, planned purchases, etc.)
2. Avoid using the flow builder or custom tools to allow maximum flexibility

Testing this agent reveals both strengths and limitations:

• It asks relevant follow-up questions to gather comprehensive financial data
• The lack of structure allows it to adapt to different user situations
• However, the output format is inconsistent, and some calculations may be inaccurate

This example highlights how some complex tasks may require additional development or integration with external tools to produce reliable, structured results.

Pricing and Limitations

Pricing Tiers

Relevance AI offers several pricing tiers:

• Free Plan: 100 credits per day
• Pro Plan: $19/month for 10,000 credits
• Team Plan: $199/month for 100,000 credits
• Enterprise: Custom pricing for higher volumes

The pricing is generally competitive with similar no-code AI platforms. Most users actively developing or deploying agents will likely need at least the Pro plan.

Key Limitations

1. Limited Integrations: While improving, the range of native integrations is somewhat limited. Some common tools and services are not yet supported.

2. Deployment Options: The lack of extensive customization for the agent interface may be limiting for customer-facing applications.

Who Should Use Relevance AI?

Relevance AI is particularly well-suited for:

• AI agencies and consultancies needing to rapidly prototype and deploy agent systems
• Businesses looking to build internal AI tools and assistants
• Developers wanting to experiment with multi-agent systems without extensive coding

It may be less ideal for:

• Projects requiring highly customized user interfaces
• Applications needing extensive integration with niche or specialized tools
• Scenarios demanding fine-grained control over agent behavior and output

Conclusion

Relevance AI offers a powerful and user-friendly platform for building AI agents and multi-agent systems. Its no-code approach and support for multi-agent collaboration make it stand out in the rapidly evolving landscape of AI development tools.

While it has some limitations in terms of customization and integrations, Relevance AI excels at enabling rapid prototyping and deployment of sophisticated AI agents. For many use cases, particularly in business and consulting contexts, it offers an excellent balance of power and ease of use.

As the platform continues to evolve, addressing current limitations around integrations and deployment options could make it an even more compelling choice for a wider range of AI development projects.

Ultimately, Relevance AI represents a significant step forward in making advanced AI agent development accessible to a broader audience, potentially accelerating innovation in this exciting field.

Article created from: https://www.youtube.com/watch?v=ig0Cu_9WmiI&t=14s