Examples & Use Cases

apiVersion: neuronetes.io/v1alpha1
kind: Model
metadata:
  name: llama-3-8b-chat
  namespace: default
spec:
  weightsURI: s3://models/llama-3-8b-instruct/
  size: 16Gi
  quantization: int8
  architecture: llama
  parameterCount: "8B"
  format: safetensors
  cachePolicy:
    priority: medium
    pinDuration: 1h
    evictionPolicy: lru

apiVersion: neuronetes.io/v1alpha1
kind: AgentClass
metadata:
  name: chat-assistant
  namespace: default
spec:
  modelRef:
    name: llama-3-8b-chat
  
  maxContextLength: 8192
  
  systemPrompt: |
    You are a helpful, friendly AI assistant. Provide clear,
    concise, and accurate responses. Be conversational but professional.
  
  temperature: 0.8
  maxTokens: 2048
  
  guardrails:
    - type: pii-detection
      action: redact
      threshold: 0.8
    - type: safety-check
      action: block
      threshold: 0.9
  
  slo:
    ttft: 300ms
    tokensPerSecond: 75
    p95Latency: 1s
  
  memoryConfig:
    type: ephemeral
    ttl: 30m
    maxSize: 5000

apiVersion: neuronetes.io/v1alpha1
kind: AgentPool
metadata:
  name: chat-pool
  namespace: default
spec:
  agentClassRef:
    name: chat-assistant
  
  minReplicas: 3
  maxReplicas: 30
  prewarmPercent: 30
  
  migProfile: "1g.5gb"
  
  autoscaling:
    metrics:
      - type: concurrent-sessions
        target: "100"
      - type: ttft-p95
        target: "300ms"
    
    behavior:
      scaleUp:
        stabilizationWindow: 30s
        maxChangeAbsolute: 5
      scaleDown:
        stabilizationWindow: 180s
        maxChangeAbsolute: 2
  
  sessionAffinity:
    enabled: true
    keyHeader: "X-Session-ID"
    ttl: 30m
    type: conversation-id
  
  scheduling:
    costOptimization:
      enabled: true
      spotEnabled: true
      maxCostPerHour: 30.0

# Get the endpoint
ENDPOINT=$(kubectl get svc neuronetes-ingress -o jsonpath='{.status.loadBalancer.ingress[0].ip}')

# Send a request
curl -X POST http://$ENDPOINT/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-Session-ID: test-session-123" \
  -d '{
    "messages": [
      {
        "role": "user",
        "content": "Hello! Can you help me understand quantum computing?"
      }
    ],
    "stream": false
  }'

import requests

def chat(message, session_id="default"):
    response = requests.post(
        f"http://{ENDPOINT}/v1/chat/completions",
        headers={
            "Content-Type": "application/json",
            "X-Session-ID": session_id
        },
        json={
            "messages": [{"role": "user", "content": message}],
            "stream": False
        }
    )
    return response.json()

# Usage
result = chat("What is the weather like today?", "user-123")
print(result["choices"][0]["message"]["content"])

apiVersion: neuronetes.io/v1alpha1
kind: AgentClass
metadata:
  name: code-assistant
spec:
  modelRef:
    name: llama-3-70b
  
  maxContextLength: 128000
  
  systemPrompt: |
    You are an expert code assistant. Help developers with:
    - Code review and suggestions
    - Bug fixing and debugging
    - Documentation generation
    - Code search and navigation
    
    Use the available tools to access the codebase.
  
  toolPermissions:
    - name: code-search
      rateLimit: 100/min
      timeout: 5s
      maxConcurrency: 3
      requiredScopes:
        - read:code
    
    - name: file-read
      rateLimit: 50/min
      timeout: 2s
      maxConcurrency: 2
      requiredScopes:
        - read:files
  
  guardrails:
    - type: pii-detection
      action: redact
    - type: code-injection-check
      action: block
  
  slo:
    ttft: 500ms
    tokensPerSecond: 50
    p95Latency: 2s

apiVersion: neuronetes.io/v1alpha1
kind: AgentClass
metadata:
  name: rag-agent
  namespace: rag-system
spec:
  modelRef:
    name: llama-3-70b-rag
  
  maxContextLength: 128000
  
  systemPrompt: |
    You are a helpful assistant with access to a knowledge base.
    
    When answering questions:
    1. Use the retrieve tool to find relevant information
    2. Cite sources from the retrieved documents
    3. Be accurate and admit when you don't know
    4. Synthesize information from multiple sources
    
    Format citations as [Source: document_id].
  
  toolPermissions:
    - name: retrieve
      rateLimit: "50/min"
      timeout: 5s
      maxConcurrency: 3
      config:
        top_k: 5
        rerank: true
        cache_ttl: 300
    
    - name: rerank
      rateLimit: "30/min"
      timeout: 2s
  
  slo:
    ttft: 800ms
    tokensPerSecond: 45
    p95Latency: 3s

apiVersion: neuronetes.io/v1alpha1
kind: AgentPool
metadata:
  name: rag-pool
  namespace: rag-system
spec:
  agentClassRef:
    name: rag-agent
  
  minReplicas: 3
  maxReplicas: 15
  
  scheduling:
    priority: 90
    
    # Co-locate with vector store for low latency
    dataLocality:
      vectorStoreAffinity:
        - weaviate
      cacheAffinity:
        - redis
    
    costOptimization:
      enabled: true
      spotEnabled: true
      sloHeadroomMs: 1500

import requests

def rag_query(question, session_id=None):
    """Query the RAG system"""
    
    response = requests.post(
        "http://rag-http.rag-system/v1/rag/query",
        headers={
            "Content-Type": "application/json",
            "X-Session-ID": session_id or "default"
        },
        json={
            "messages": [
                {
                    "role": "user",
                    "content": question
                }
            ],
            "stream": False,
            "retrieval_config": {
                "top_k": 5,
                "rerank": True
            }
        }
    )
    
    return response.json()

# Example usage
result = rag_query(
    "What are the best practices for Kubernetes security?",
    session_id="user-123"
)

print(result["choices"][0]["message"]["content"])
print(f"Sources used: {result['metadata']['sources_used']}")