Frequently Asked Questions
NVIDIA AgentIQ
Q: What is AgentIQ?
- A lightweight, unifying library that makes every agent, tool, and workflow you already have work together, just as simple function calls work together in complex software applications.
- An end-to-end agentic profiler, allowing you to track input/output tokens and timings at a granular level for every tool and agent, regardless of the amount of nesting.
- A way to accomplish end-to-end evaluation and observability. With the potential to wrap and hook into every function call, AgentIQ can output observability data to your platform of choice. It also includes an end-to-end evaluation system, allowing you to consistently evaluate your complex, multi-framework workflows in the exact same way as you develop and deploy them.
- A compliment to existing agentic frameworks and memory tools, not a replacement.
- 100% opt in. While we encourage users to wrap (decorate) every tool and agent to get the most out of the profiler, you have the freedom to integrate to whatever level you want - tool level, agent level, or entire workflow level. You have the freedom to start small and where you believe you’ll see the most value and expand from there.
Q: Why should developers use AgentIQ?
AgentIQ is a flexible library that allows for easy connection of existing enterprise agents, across any framework, to data sources and tools. The core principle of this library is every agent, tool, and agentic workflow exists as a function call - enabling composability between these agents, tools, and workflows that allow developers to build once and reuse in different scenarios. This makes AgentIQ able to work across any agentic framework, combining existing development work and reducing the need to replatform. This library is agentic framework agnostic, long term memory, and data source agnostic. It also allows development teams to move quickly if they already are developing with agents- focusing on what framework best meets their needs, while providing a holistic approach to evaluation and observability. A core component of AgentIQ is the profiler, which can be run to uncover hidden latencies and suboptimal models/tools for specific, granular parts of pipelines. An evaluation system is provided to help users verify and maintain the accuracy of the RAG and E2E system configurations.
Q: What is the value of the NVIDIA AgentIQ library?
With the NVIDIA AgentIQ open-source toolkit, developers across the organization have access to a collection of reusable, best-in-class tools, pipelines, and workflows to build agentic systems. Using universal descriptors for agents, tools, and workflows, developers can flexibly choose agent frameworks best suited to each task in a workflow and connect them all easily. Fine-grained telemetry provided by the AgentIQ software can be used to accelerate agentic systems, enabling better business outcomes with existing infrastructure. The evaluation capabilities of the AgentIQ library improves the accuracy of agentic systems, uncovering areas for improvement and providing recommendations.
Q: When will the AgentIQ framework be available? How can developers use it?
After the NVIDIA GTC keynote, March 18, 2025, the open-source AgentIQ library will be available on GitHub.
Q: How much does the AgentIQ framework cost?
The AgentIQ framework is free.
Q: How does AgentIQ compare to other available agentic frameworks and platforms?
- An agentic framework. AgentIQ is built to work side-by-side and around existing agentic frameworks, including LangChain, Llama Index, Crew.ai, Microsoft Semantic Kernel, MCP, and many more - including customer enterprise frameworks and simple Python agents.
- An attempt to solve agent-to-agent communication. Agent communication is best handled over existing protocols, such as HTTP, gRPC, and sockets.
- An observability platform. While AgentIQ is able to collect and transmit fine-grained telemetry to help with optimization and evaluation, it does not replace your preferred observability platform and data collection application.
Q: Which agent frameworks does the AgentIQ library work with?
AgentIQ seamlessly integrates with frameworks like CrewAI, LangGraph, Llama Stack, Microsoft Azure AI Agent Service, Anthropic’s Model Context Protocol and Letta to let developers work seamlessly in their preferred environment. Microsoft Azure AI Agent Service is integrated with AgentIQ to enable more efficient AI agents and orchestration of multi-agent frameworks using Semantic Kernel, which is fully supported in AgentIQ.
Q: How does AgentIQ compare with existing agent building frameworks and agent orchestration tools available today?
The goal with AgentIQ is to provide an open source library that developers can both use and contribute to, which is interoperable with any agent framework or orchestration tool used by enterprises. With AgentIQ organizations have access to telemetry that can be used to accelerate the performance of the entire agentic AI system—achieving faster responses to agent tasks, and developer velocity is also accelerated as they can do more work, faster. AgentIQ also enables developers to evaluate agentic systems and quickly and easily get recommendations for ways to improve.
Q: How does AgentIQ compare with Model-Context Protocol (MCP)
AgentIQ integrates with Anthropic’s MCP so that developers can use AgentIQ seamlessly with their preferred tools in their existing environment.
Q: What languages are supported for integrating agents into AgentIQ?
AgentIQ supports Python.
Q: What are the future enhancements planned for AgentIQ?
- Additional end-to-end agentic use case for software engineering (Q2 2025)
- AI-Q NVIDIA Blueprint customization guide (Q2 2025)
- Optional integration of NVIDIA NeMo Guardrails (Q2 2025)
- Agentic system level accelerations via integrations with NVIDIA Dynamo (Q3 2025)
- Flexible deployment and workflow instantiation option (Q4 2025)
- Addition of data feedback and improvement loop (data flywheel) (Q4 2025)
The AgentIQ library will enable NVIDIA to implement and orchestrate system level accelerations that require control of placement and configuration of agentic system components.
Q: Where can I find more information about how to use AgentIQ?
See the AgentIQ documentation here.