Automated governance audits of the most popular open-source AI-agent frameworks — scored on 17 dimensions across 12 scan layers.
What is this? Warden is an open-source CLI scanner that evaluates
AI-agent projects across 17 governance dimensions: tool-call enforcement,
agent identity, audit trails, credential management, supply-chain, trap defense,
and more. Every project below was scanned with the latest version of Warden,
producing the same HTML report you'd get locally by running
pip install warden-ai && warden scan <path>.
What do the scores mean? These scores reflect the governance posture of the framework's own source code, not of applications built on top of it. Most general-purpose frameworks land in AT_RISK or PARTIAL territory — that's expected. They are libraries, not governance platforms. If your team needs full gateway-level enforcement, you probably want a dedicated runtime layer on top.
| Project | Category | Description | Score | Level | Critical | |
|---|---|---|---|---|---|---|
| VigIA Orchestrator | Framework | C#/.NET agent orchestrator built on Microsoft.Extensions.AI with Result<T,E>, invariant enforcement, and FSM-guarded state transitions. The first non-Python project in the gallery. | 61/100 | PARTIAL | 0 | Audit → |
| PydanticAI | Framework | Agent framework from the Pydantic team, designed for type-safe LLM applications. | 24/100 | UNGOVERNED | 17 | Audit → |
| CrewAI | Multi-Agent | Framework for orchestrating role-playing autonomous AI agents that collaborate on tasks. | 19/100 | UNGOVERNED | 21 | Audit → |
| Langflow | Low-Code | Visual low-code tool for building and deploying AI agents and workflows. | 18/100 | UNGOVERNED | 21 | Audit → |
| Haystack | RAG | End-to-end framework for building production-ready LLM and RAG applications. | 15/100 | UNGOVERNED | 45 | Audit → |
| LangGraph | Orchestrator | Graph-based orchestration for multi-agent LLM workflows by the LangChain team. | 14/100 | UNGOVERNED | 4 | Audit → |
| Semantic Kernel | Framework | Microsoft's open-source SDK for integrating LLMs into conventional applications. | 14/100 | UNGOVERNED | 15 | Audit → |
| LangChain | Framework | The most-starred framework for building LLM applications with chains, agents, and tools. | 13/100 | UNGOVERNED | 3 | Audit → |
| LlamaIndex | RAG | Data framework for connecting custom data sources to LLMs, the standard for RAG pipelines. | 13/100 | UNGOVERNED | 13 | Audit → |
| MetaGPT | Multi-Agent | Multi-agent framework that assigns different roles to GPT agents to collaborate on complex tasks. | 11/100 | UNGOVERNED | 11 | Audit → |
| AutoGen | Multi-Agent | Microsoft's framework for building multi-agent conversational AI applications. | 6/100 | UNGOVERNED | 44 | Audit → |
pip install warden-ai
warden scan /path/to/your/projectZero telemetry, zero cloud calls. Everything runs locally. See github.com/SharkRouter/warden for the source, scoring model, and methodology.