aasimmalikin/qagent

Qagent is a Python project that automatically drafts answers to vendor security questionnaires, the long lists of questions a company asks a supplier about how it protects data. It is also written as a teaching example, showing exactly how a multi agent AI pipeline works under the hood by building all of the concurrency and reliability plumbing by hand instead of hiding it inside a framework like LangChain or CrewAI. The only third party libraries used are httpx and pydantic. By default, the project runs completely offline with no API key at all, since the AI model is simulated with realistic delays and occasional failures, so anyone can clone it and immediately watch how it behaves. Each security question goes through its own small loop: it looks up relevant information, drafts an answer, scores how confident it is, and tries again with a wider search if the confidence is too low. Many questions are processed at the same time through a worker pool, with built in safeguards such as limiting how many requests run at once, retrying failed calls with random delays, timing out calls that hang, and isolating a single failing question so it does not stop the rest of the batch from finishing. To run it, you need Python 3.11 or newer, since it relies on newer asyncio features. After cloning the repository and setting up a virtual environment, you install the two dependencies and run the project as a Python module rather than running the main file directly, since it uses relative imports internally. Running it processes fourteen sample questions and writes a results.json file summarizing how each one was answered, including any that were flagged for a human to review. The project can also be pointed at a real AI provider instead of the simulator by editing two specific files, without changing any of the surrounding reliability code. The README does not state a license for this project.