RadOps is an AI-driven multi-agent platform positioned for NetDevOps and security automation, with the goal of automating DevOps workflows with near-human reasoning capabilities. Unlike traditional chatbots, RadOps remembers context, verifies its own work, and can autonomously execute complex multi-step operations across the entire infrastructure.
Purpose and Positioning RadOps is primarily aimed at network operations (NetDevOps) and security automation scenarios. It orchestrates and executes business processes through multi-agent collaboration, making it suitable for teams that need to perform multi-step operations across infrastructure, such as network configuration, security checks, and other automation tasks.
Main Capabilities The platform uses a Supervisor-Worker architecture and employs strict sequential logic for constrained orchestration to prevent execution errors. It features three layers of cognitive memory: working memory, temporary facts, and core knowledge (permanent architecture rules). Agent discovery supports two scalable modes: prompt mode, suitable for small teams (fewer than 15 agents), and discovery mode, which enables unlimited scaling through vector-based agent lookup.
Configuration-driven expert agents allow specialized agents (such as networking and security) to be created quickly by defining roles and tool sets in YAML, without writing new code. Human-in-the-loop functionality can pause workflows before executing sensitive operations and wait for user approval or input. Agent skills can quickly extend an agent's capabilities through Markdown files, and skills can be Python, Bash, or MCP calls. Unified workflows support defining complex standard operating procedures (SOPs) in Markdown, orchestrating tools, skills, and MCP calls, and supporting variable injection. Multi-step workflows automatically break complex requests into logical steps, execute them in order, track status, and enforce the plan.
Verification Mechanism The platform provides a "trust but verify" audit mechanism, in which dedicated QA audit nodes validate actual tool output against the user request, catching hallucinations before they reach the user.
Data and Connectivity RadOps adopts declarative RAG and BYODB (bring your own database), supports mainstream vector databases, and generates knowledge tools in a zero-code, configuration-driven way. It builds resilient connections based on the Model Context Protocol (MCP), with a self-healing client that can continue working after a server restart. It provides complete tracing of agent logic, tool execution, and LLM streaming output through OpenTelemetry, enabling deep observability.
Supported ModelsProviders OpenAI (openai): cloud models such as gpt-5 and gpt-5-nano. Anthropic (anthropic): cloud models such as claude-4-5-sonnet and claude-4-5-opus. DeepSeek (deepseek): DeepSeek API models. Azure OpenAI (azure): OpenAI models hosted on Azure. Google (google): Google Gemini models such as gemini-3-pro-preview. Groq (groq): Groq Cloud models. Mistral (mistral): Mistral AI models. AWS Bedrock (bedrock): AWS-hosted models. Ollama (ollama): local models. If used for agents, the model must support tool calling.
Vector databases Weaviate: hybrid search, GraphQL API, multi-tenancy. Qdrant: high performance (Rust), advanced filtering. Pinecone: managed cloud, serverless, auto-scaling. Milvus: open source, horizontal scaling, GPU support. Chroma: lightweight, embedded, suitable for development/testing.
How to use You can install by cloning the repository: git clone https://github.com/mehrdadrad/radops.git，然后 cd radops, install dependencies with uv: uv pip install -e . . For more detailed configuration, deployment, and feature guides, please refer to the official documentation. The project welcomes contributions; the steps are: fork the project on GitHub, create a feature branch (git checkout -b feature/amazing-feature), commit your changes, push the branch, and open a Pull Request.
Built with RadOps is built with LangGraph, Mem0, mainstream vector databases, and more.
Use cases and limitations RadOps is suitable for teams that need to automate network and security operations processes. Its limitation is that if Ollama local models are used to run agents, the model must support tool calling; other specific limitations are not provided at this time.
