In June 2026 the IETF published RFC 10008, which defines a brand new HTTP method: QUERY. It is the first new general-purpose method since PATCH (RFC 5789, 2010), and it closes a gap that has existed for as long as HTTP itself: performing a safe, idempotent request that carries a request body. In this post we look at the method registry as it stands today, what exactly QUERY adds, how it interacts with caching and conditional requests, and what the support story looks like in OpenAPI and FastAPI.

In the previous post, we built a dev.bg job-search agent with Agno and MLX that supports Human-in-the-Loop (HITL) control flow, meaning the agent pauses when it needs missing fields, asks for structured user input, and resumes from exactly where it left off. But there’s a recurring question in any agent project: How do we know the system prompt is actually good? You write something that works in your head, you tweak it a few times, and you ship it. But you’re essentially doing manual trial-and-error on natural language. In this post, we’ll do something more principled: use Microsoft’s Agent Lightning (agentlightning) to automatically optimize the system prompt for the same job-search agent. Instead of guessing the best prompt, we will let the model learn it.

In my earlier posts, we first built a dev.bg job-search assistant by serving a local MLX model via mlx_lm.server, then using the OpenAI Python client to do tool calling. The client code owned the whole “tool loop”: call model → execute tool → append tool result → call model again, etc: A Job Postings Tool: A Guide to MLX-LM Server and Tool Use with the OpenAI Client. Later, we connected MLX to the Agno framework using theOpenAILike model adapter, and we saw how easy it is to give an Agno agent tools: Running Local Hugging Face Models with MLX-LM and the Agno Agentic Framework. In this post, we’ll rebuild the same agent with Agno and MLX, but this time we’ll focus on Human-in-the-Loop (HITL) control flow. Concretely, the agent can pause when it needs missing info, request that info via a structured user input (schema), continue from exactly where it left off.

This post covers the technical and theoretical differences between Edge Functions and traditional Serverless Functions (like AWS Lambda). We are going to explore how they differ in runtime architecture, performance characteristics, and show 2 simple use cases.

In this blog post we are going to show how to run a Hugging Face model locally using Apple's MLX-LM framework, connect it to the Agno agentic framework, and extend the agent's capabilities with custom tools. Agno is a modern and extensible agentic framework for Python. It is designed to be model-agnostic, allowing to seamlessly switch between different LLM providers and models. Medium post can be found here.