pyspur-dev/pyspur

PySpur is a visual tool for building and testing AI agents, aimed at engineers who spend too much time tweaking prompts and staring at raw terminal output to figure out why an AI workflow is misbehaving. The idea is to give you a canvas where you can see each step of an AI workflow as a connected block, run it, inspect what happened at each node, and iterate quickly without writing a lot of glue code from scratch. The workflow is organized around four steps: define test cases, build the agent using either a Python file or the visual editor, run it repeatedly while adjusting things, and then deploy it as an API. Each of those steps has a corresponding view in the interface. You can run the same test case across multiple versions of a workflow side by side to compare outputs, which is useful when deciding whether a prompt change actually improved results. Feature-wise, PySpur supports pausing a workflow mid-run to wait for a human to review and approve output before continuing. It can process files, PDFs, videos, images, and audio. It includes tools for connecting to external services like Slack, GitHub, and Google Sheets. It also supports RAG workflows, which means you can feed it a collection of documents, have it chunk and index them, and then retrieve relevant pieces during a conversation or task. It connects to over 100 AI model providers, so you are not locked into a single vendor. Installing it locally takes three commands: pip install pyspur, initialize a project, and start the server. A local interface then opens at port 6080. The project recommends PostgreSQL for production stability but ships with SQLite for quick evaluation. PySpur is open source and written in Python for the backend logic, with a TypeScript frontend. Development setup instructions cover Unix-like systems only, Windows is not supported.