What is cybnodes?

CybNodes is a small Python library, installed with pip, that wraps around any language model you already use and surrounds it with networks of verifiable capabilities: exact calculation, stored knowledge, and web search. Rather than making the model itself bigger or smarter, the idea is to place reliable, checkable tools in front of it, and have those tools' raw answers rephrased back in the voice of your own model, so the answer sounds like your assistant while staying factually exact. The problem it addresses is straightforward: language models are good at conversing but bad at guaranteeing correctness. They might get simple arithmetic wrong, invent a date, or lack current information. CybNodes routes a question to a specialized network first, and only lets the model answer directly when none of the networks recognize the question. A router decides which network, if any, should handle a given question, and a component the author calls a weaver restates the network's raw result in the model's own voice, so the personality carries through even for a purely mechanical answer. The library has no required external dependencies beyond Python's own standard library, and works with Python 3.9 or newer. You bring your own model as any callable function, whether that is a local model, a hosted API, or something else. Several ready made networks ship with the library. One does exact arithmetic safely, without a raw code evaluator. Another does symbolic math like derivatives and integrals, using the optional sympy library. Another answers from a small knowledge graph you provide. Another recalls previously validated question and answer pairs word for word rather than paraphrasing them, to avoid introducing errors. A wrapping component checks whether a network's answer is supported by nearby stored facts before allowing it through. The last one performs a live web search using the Brave Search API and returns a link to its source. Writing a new custom network is meant to be simple: you define a method that decides whether a given question belongs to your network, and if so, produce a result with both the answer and a real, checkable source for it. The project also includes an optional manifest system where each network declares what kinds of questions it answers, meant to help with documentation and future routing. The project is licensed under MIT and available on PyPI.