PySpur

PySpur is an open-source framework for building, testing, and deploying AI agents. It pairs a visual drag-and-drop workflow editor with first-class Python so engineers can sketch a pipeline graphically, then drop into code for custom tools, evaluators, or nodes. Agents are exported as JSON, which makes them easy to diff, code-review, and roll out via the same git flow you already use for the rest of your stack.

The differentiator is the emphasis on the iteration loop. Most agent frameworks (LangGraph, CrewAI, etc.) leave you to bolt on your own test rig; PySpur ships test-case definition, batch evaluation, and failure inspection in the core product, which is closer in spirit to LangSmith than to a pure orchestration library. It's Apache-2.0 licensed with 5,500+ GitHub stars, installs via pip, and runs equally well on a laptop, a cloud VM, or your own hardware. A managed Cloud tier and Python/TypeScript SDKs are flagged as forthcoming.

PySpur is model-agnostic: you wire up whichever LLM provider the node calls, so you can mix Claude, GPT, open-weight models, or local Ollama within one graph. It's a sensible pick for teams that want a self-hosted alternative to closed visual builders like Flowise Cloud or Relevance AI, particularly where the agent graph needs to live next to the rest of your application code.