Qat — your AI agent for quantum computing

About this project
An AI agent that takes a quantum-computing experiment from a question to a result you can trust: it plans, computes with code, checks against the exact answer, and asks before anything risky.
A quantum experiment is a chain of manual steps: read the documentation, build the Hamiltonian, design the circuit, choose an optimiser, simulate, compare with the exact answer, write it up. A general chatbot describes these steps but does not run or check them, and a fluent answer with a wrong number looks exactly like a right one.
This agent is built around one rule: never let the model grade its own work. The model writes a recipe; code builds the circuit and computes every number; each result is compared with an exact reference; a human approves generated code, memories, web sources and real quantum-computer time.
What it does: explains concepts from a knowledge base with citations; solves molecules (VQE, QPE), optimisation problems (QAOA), factoring (Shor) and differential equations, each against an exact reference; makes circuits smaller with proof of equivalence; runs circuits on a noisy device or a quantum computer and ranks error-mitigation settings against the simulator; improves a result in loops until a measured goal is met; researches papers and builds new tools in a sandbox.
Measured: 15 of 15 tasks verified with no false passes, 17 of 18 tool choices right, RAGAs faithfulness 0.85, 11 of 11 attack prompts handled; harness changes took the whole agent from 13/15 to 15/15 checked cases at a third of the cost.
Tech: LangChain v1 and LangGraph, Chroma, Streamlit, PennyLane, PySCF, Qiskit, OpenRouter and Gemini models.