deepteam
View on GitHubDeepTeam is a framework to red team LLMs and AI agents.
DeepTeam is an open-source Python framework for red teaming LLM apps and AI agents. It ships 50+ vulnerability checks and 20+ adversarial attack methods, runs locally on top of DeepEval, and adds optional guardrails for production use.
Use Cases
Red teaming LLM chatbots and AI agentsSimulating jailbreaks and prompt injection attacksTesting RAG pipelines for PII leakage and cross-context retrievalAuditing agents for excessive agency and tool misuseRunning bias, toxicity and fairness evaluationsAdding LLM security tests to CI/CD pipelinesMeeting OWASP LLM Top 10 / compliance requirementsBenchmarking vulnerabilities with custom criteriaComparing model safety before and after guardrailsPenetration testing code assistants and IDE agents
Built With
- Language
- Python
- Frameworks
- DeepEval · OpenAI SDK · Claude Agent SDK · Cursor SDK · OpenAI Codex · aiohttp · Poetry
Tags
red-teaming · llm-security · ai-safety · guardrails · jailbreak · prompt-injection · llm-evaluation · adversarial-attacks · vulnerability-scanning · ai-agents · penetration-testing · bias-detection · multi-turn · python · llm-testing · responsible-ai