ai-pi-nv-nemotron-ultra

name: ai-pi-nv-nemotron-ultra description: Adversarial review via pi using Llama 3.1 Nemotron Ultra (253B) via nvidia-nim version: 1.0.0 enforcement: strict triggers: - /ai-pi-nemotron-ultra - /pi-nemotron-ultra - adversarial review nemotron-ultra workflow_steps: - Parse target - Invoke pi with nemotron ultra via nvidia-nim - Parse JSON output - Report findings allowed_tools: - Bash - Read

/ai-pi-nemotron-ultra — Adversarial Review via Pi + Llama 3.1 Nemotron Ultra

Quick Start

pi --model nvidia-nim/nvidia/llama-3.1-nemotron-ultra-253b-v1 \
  -p @P:/path/to/file.py \
  "Read the file at @{path} carefully. Return JSON with score (0-1), summary (1 sentence), and issues. Begin your response with the exact first line of the file, quoted. If you did not read the file, set score to 0.0."

Model Info

API Key Setup

pi requires NVIDIA_NIM_API_KEY at startup. Export from auth.json:

NV_KEY=$(python -c "import json; d=json.load(open('$USERPROFILE/.pi/agent/auth.json')); print(d.get('nvidia',{}).get('key',''))")
export NVIDIA_NIM_API_KEY="$NV_KEY"
pi --model nvidia-nim/nvidia/llama-3.1-nemotron-ultra-253b-v1 -p @P:/path/to/file.py "Read the file at @{path} carefully. Return JSON with score (0-1), summary (1 sentence), and issues. Begin your response with the exact first line of the file, quoted. If you did not read the file, set score to 0.0."

Verification

pi --model nvidia-nim/nvidia/llama-3.1-nemotron-ultra-253b-v1 -p @P:/path/to/file.py "Say hello"

If you get 429 status code — rate limit hit. Wait 30s and retry.