PySpur

PySpur lets an AI engineer define test cases, build an agent visually or in Python code, and iterate on it before deployment, aiming to reduce the trial-and-error of prompt tweaking and the difficulty of debugging raw agent output.

It supports persistent human-in-the-loop workflows that wait for approval, iterative tool-calling loops with memory, file upload and URL pasting for document processing, a UI editor for structured JSON Schema outputs, retrieval-augmented generation (parsing, chunking, embedding and upserting data into a vector database), multimodal input across video, image, audio, text and code, built-in tool integrations such as Slack, Firecrawl, Google Sheets and GitHub, execution traces, evaluation against real-world datasets, and one-click deployment as an API.

It installs with pip install pyspur, requires Python 3.11 or higher, is written in Python, and supports more than 100 LLM providers, embedders and vector databases.