Safety Guardrails for Large Language Model Workflows // A model- agnostic safety layer for developers building with large language model (LLM) applications. Maps starter controls to the Open Worldwide Application Security Project Top 10 for Large Language Model Applications 2025 risk categories https://genai.owasp.org/llm-top-10/ via a modular rule engine. Supports regular-expression rules, lightweight natural language processing (NLP) intent checks, optional scanners, and semantic large language model reviewer checks on prompts, conversations, retrieved context, tool inputs and outputs, streaming chunks, and model outputs. Supports workflows with the 'Ollama' local web service https://ollama.com/ via 'ellmer', remote reviewer endpoints, and other chat interfaces callable from 'R'. Intended as an experimental guardrail layer that teams should evaluate against their own workflows before relying on it in production.