ArchGateway Explained AI Native Infrastructure for Agentic Workflows

Key Takeaways

[Music] We talked about now a product which is part of the component in building agentic AI applications. Arch gateway an AI native edge and LLM gateway server or agentic apps. The essence is you have application code and then you have infrastructure and tools and infrastructure tools like vector databases and memory APIs tend to handle the pesky heavy lifting so you can move quickly and essentially this is where it sits in the stack. You have a human chatting with sub interface. Arch gateway with purpose-designed fast large language models does two tasks. Clarifies the question on behalf of the user based on the configuration provided. Extract structured data information like entities relations from the user query and passes that to your application server and asks it to complete the task. Once it reads receives a response based on some on convenience features, it can simply dispatch the summarization call to one or many more large language models configured using a unified interface and starts logging metrics and traces in metric stores like Jagger, signals, honeycomb addition any other open telemetry based so that you can start moving faster in thinking and building about agentic scenarios. The more future work that we're doing at the moment is supporting full autonomous scenarios where essentially gate that large gateway becomes a router to different autonomous agents does quick task clarification questions forward presents to the agent the full query. Just think of it as a TA to a professor in that scenario and then be able to do um uh effectively handle these things outside of application code. So you're focusing on more high level objectives, not the nitty-gritty uh pesky details of prompt engineering, prompt user query parsing, intent detection, uh routing, etc.