In the rapidly evolving landscape of artificial intelligence, AI agents are becoming increasingly prevalent. While the technology itself is readily available, it's the developers who are driving its integration into enterprise environments. Although no-code tools have democratized access for non-technical users, we recognize that a more nuanced approach is needed – one that caters to both technical and non-technical users alike.
Agent Building Frameworks
One of the earlier frameworks for agent building was popularized by LangChain. The fundamental approach to build agents is to provide LLMs with access to tools like search capabilities and database queries. This innovation sparked a wave of development, leading to the emergence of numerous frameworks including LangGraph, LlamaIndex, OpenAI's Swarm, CrewAI, Microsoft AutoGen, IBM's Bee Agent Framework and others. Essentially, these frameworks are wrappers for Python and LLM providers such as OpenAI and Anthropic, enabling the development of agents and multi-agent-systems.
The Challenge of Deploying Agents
Despite these advances, a significant hurdle remains: the complexity and cost of hosting agents and managing their operations. Many developers and organisations struggle with the infrastructure requirements and financial implications of maintaining in-house or externally-facing agents. Costs of third party APIs used in tools such as for search capabilities (Google Search, Apify) and data enrichment (Apollo, LinkedIn) also present significant costs in situations where usage is less well-defined.
This challenge mirrors the early days of automation tools, before platforms like Zapier and Make streamlined the process of cloud hosting and credit-based usage for API integrations. We believe the same can be replicated with the agent building experience.
Bridging the Gap with Developer-First Tools
Relevance’s Python SDK can enable developers to build robust, production-ready agents, while our platforms flexibility allows for deep customization while maintaining the ease of use that developers expect from modern tools.
The python SDK for RelevanceAI is available here: https://github.com/RelevanceAI/relevanceai
We’re addressing these challenges to improve the agent building experience:
- Cloud-Based Agent Execution via API: Our cloud infrastructure eliminates the complexities of hosting and managing agents locally, providing a scalable and cost-effective solution.
- Access to Third Party Providers: Unlike many existing solutions, we've simplified access to third-party providers, creating a more integrated and comprehensive development environment. Usage is credit-based, allowing teams and agent-based products to scale economically.
- Hybrid Approach to Accessibility: We maintain the benefits of low-code platforms for non-technical users while providing powerful SDK capabilities for developers.
Real-World Applications
Our examples folder of the Python SDK showcase some practical applications in both application development and data science workflows. We’re aiming to provide developers with concrete implementations and best practices for integrating AI agents into their projects.
See an example below of where agents can be easily integrated within apps and other internal tools. Below is a form application built on Streamlit can integrate a Relevance AI agent to handle research and qualification before inputting the results into an AirTable database. Attach Relevance AI agents within your application backend with a few lines of code.
You can find an example here: https://github.com/ethantrang/qualification-app
Looking forward
As AI agents continue to evolve, the need for developer-friendly tools becomes increasingly crucial. By combining the power of SDK-based development and a cloud-based platform, Relevance AI wants to make sophisticated AI agent development accessible to teams of all sizes and technical capabilities.
