aegis_v4_safety: resources_servers: aegis_v4_safety: entrypoint: app.py domain: safety verified: false description: Multimodal response-safety evaluation with NVIDIA Nemotron 3 Content Safety (Aegis v4) value: Classify target-model prompts and responses as safe or unsafe and report safety categories judge_model_server: type: responses_api_models name: aegis_v4_model judge_responses_create_params: input: [] max_output_tokens: 200 temperature: 0.01 top_p: 0.95 judge_endpoint_max_concurrency: 64 response_text_mode: final_only reward_if_safe: 1.0 reward_if_unsafe: 0.0 reward_if_unresolved: 0.0 aegis_v4_safety_simple_agent: responses_api_agents: simple_agent: entrypoint: app.py max_steps: 1 resources_server: type: resources_servers name: aegis_v4_safety model_server: type: responses_api_models name: policy_model datasets: - name: example type: example jsonl_fpath: resources_servers/aegis_v4_safety/data/example.jsonl num_repeats: 1 license: Apache 2.0 aegis_v4_safety_environment_server: environment_servers: single_agent_turn_legacy: entrypoint: app.py resources_server: type: resources_servers name: aegis_v4_safety agent_server: type: responses_api_agents name: aegis_v4_safety_simple_agent resources_tool_transports: - direct_http