An AI agent that reasons, searches, and writes its own code

Agenta is a research agent built on a LangGraph StateGraph that decides for itself which tool to reach for next. It can reason through a problem, search the web when it lacks context, execute Python to verify a calculation, and then synthesise a conclusion — looping back whenever a step comes up short.

The hard part isn't calling a model, it's managing state across a multi-step run. I built explicit graph nodes for reason, search, code, and conclude, with a FastAPI backend streaming every intermediate step to the client over Server-Sent Events so you can watch the agent think rather than staring at a spinner.