OSSA Execution Profiles

Execution profiles provide pre-configured settings for different agent execution scenarios. They control LLM parameters, validation requirements, and runtime behavior to optimize for speed, thoroughness, or safety.

Overview

OSSA v0.3.2 defines four standard execution profiles:

Profile Configuration

Basic Structure

apiVersion: ossa/v0.4.9
kind: Agent

spec:
  execution_profile:
    default: balanced
    profiles:
      fast:
        maxTokens: 4000
        temperature: 0.0
        description: "Quick triage and responses"
      balanced:
        maxTokens: 16000
        temperature: 0.1
        description: "Standard operations"
      deep:
        maxTokens: 32000
        temperature: 0.0
        reasoning_enabled: true
        description: "Deep analysis and reasoning"
      safe:
        temperature: 0.0
        validation_required: true
        audit_log: true
        description: "Compliance-grade execution"

Profile Selection

Profiles can be selected at runtime via environment variables:

spec:
  llm:
    profile: ${LLM_PROFILE:-balanced}
  
  execution_profile:
    default: ${LLM_PROFILE:-balanced}

Profile Definitions

Fast Profile

Optimized for quick responses and triage operations.

apiVersion: ossa/v0.4.9
kind: Agent

spec:
  execution_profile:
    profiles:
      fast:
        maxTokens: 4000
        temperature: 0.0
        description: "Quick triage and responses"
        
        # Reduced timeouts
        timeout_seconds: 30
        
        # Skip optional validations
        validation_required: false
        
        # Minimal reasoning
        reasoning_enabled: false
        
        # No audit overhead
        audit_log: false

Use Cases:

• Initial query classification
• Simple Q&A responses
• Status checks and health pings
• Quick data lookups

Trade-offs:

• Lower token limits may truncate complex outputs
• No extended reasoning capabilities
• Reduced validation may miss edge cases

Balanced Profile

Default profile for standard agent operations.

apiVersion: ossa/v0.4.9
kind: Agent

spec:
  execution_profile:
    profiles:
      balanced:
        maxTokens: 16000
        temperature: 0.1
        description: "Standard operations"
        
        # Standard timeouts
        timeout_seconds: 300
        
        # Basic validation
        validation_required: true
        
        # Standard reasoning
        reasoning_enabled: false
        
        # Selective audit logging
        audit_log: false

Use Cases:

• General-purpose agent tasks
• Customer support interactions
• Code assistance and reviews
• Data analysis and reporting

Trade-offs:

• Balanced between speed and thoroughness
• Suitable for most production workloads
• May need profile upgrade for complex tasks

Deep Profile

Extended capabilities for complex analysis and reasoning.

apiVersion: ossa/v0.4.9
kind: Agent

spec:
  execution_profile:
    profiles:
      deep:
        maxTokens: 32000
        temperature: 0.0
        description: "Deep analysis and reasoning"
        
        # Extended timeouts
        timeout_seconds: 600
        
        # Strict validation
        validation_required: true
        
        # Enable extended reasoning (chain-of-thought)
        reasoning_enabled: true
        
        # Optional audit for analysis
        audit_log: false
        
        # Additional deep profile settings
        max_iterations: 20
        reflection_enabled: true

Use Cases:

• Complex problem solving
• Multi-step reasoning tasks
• Threat analysis and security reviews
• Research and deep investigation
• Code architecture analysis

Trade-offs:

• Higher token costs
• Longer execution times
• More computational resources

Safe Profile

Compliance-grade execution with full audit trails.

apiVersion: ossa/v0.4.9
kind: Agent

spec:
  execution_profile:
    profiles:
      safe:
        temperature: 0.0
        description: "Compliance-grade execution"
        
        # Deterministic output
        temperature: 0.0
        top_p: 1.0
        
        # All validations required
        validation_required: true
        
        # Full audit logging
        audit_log: true
        
        # Human approval gates
        require_approval: true
        
        # Strict output validation
        output_validation: strict
        
        # No caching (fresh computation)
        cache_enabled: false

Use Cases:

• Security scanning and compliance
• Financial operations
• Healthcare and regulated industries
• Legal document processing
• Data privacy operations

Trade-offs:

• Slowest execution profile
• Highest operational overhead
• May require human-in-the-loop

Profile Configuration Options

LLM Parameters

Execution Settings

Audit and Compliance

Runtime Profile Selection

Environment Variables

# Set default profile
export LLM_PROFILE=balanced

# Override for specific runs
LLM_PROFILE=deep ossa run my-agent

Agent Manifest

apiVersion: ossa/v0.4.9
kind: Agent

spec:
  llm:
    provider: ${LLM_PROVIDER:-anthropic}
    model: ${LLM_MODEL:-claude-sonnet}
    profile: ${LLM_PROFILE:-balanced}
    
    # Profile-based parameters
    temperature: ${LLM_TEMPERATURE:-0.1}
    maxTokens: ${LLM_MAX_TOKENS:-16000}

Dynamic Profile Switching

Agents can switch profiles during execution based on task requirements:

apiVersion: ossa/v0.4.9
kind: Agent

spec:
  capabilities:
    - name: security_scan
      type: action
      execution_profile: safe  # Override for this capability
      
    - name: quick_check
      type: query
      execution_profile: fast  # Quick responses
      
    - name: deep_analysis
      type: action
      execution_profile: deep  # Complex analysis

Complete Example

A security scanner agent with profile-aware execution:

apiVersion: ossa/v0.4.9
kind: Agent

metadata:
  name: security-scanner
  version: 1.0.0
  description: Security scanner with profile-aware execution

spec:
  # LLM Configuration with Profile
  llm:
    provider: ${LLM_PROVIDER:-anthropic}
    model: ${LLM_MODEL:-claude-sonnet}
    profile: ${LLM_PROFILE:-safe}
    
    temperature: ${LLM_TEMPERATURE:-0.0}
    maxTokens: ${LLM_MAX_TOKENS:-16000}
    topP: ${LLM_TOP_P:-0.9}

    # Multi-provider fallback
    fallback_models:
      - provider: ${LLM_FALLBACK_PROVIDER_1:-openai}
        model: ${LLM_FALLBACK_MODEL_1:-gpt-4o}
        condition: on_error
      - provider: ${LLM_FALLBACK_PROVIDER_2:-google}
        model: ${LLM_FALLBACK_MODEL_2:-gemini-2.0-flash}
        condition: on_rate_limit

    retry_config:
      max_attempts: ${LLM_RETRY_ATTEMPTS:-3}
      backoff_strategy: ${LLM_BACKOFF_STRATEGY:-exponential}
      initial_delay_ms: 1000
      max_delay_ms: 30000

  # Execution Profiles
  execution_profile:
    default: ${LLM_PROFILE:-safe}
    profiles:
      fast:
        maxTokens: 4000
        temperature: 0.0
        description: "Quick triage scans"
        timeout_seconds: 30
        validation_required: false
        audit_log: false

      balanced:
        maxTokens: 16000
        temperature: 0.1
        description: "Standard security analysis"
        timeout_seconds: 300
        validation_required: true
        audit_log: false

      deep:
        maxTokens: 32000
        temperature: 0.0
        reasoning_enabled: true
        description: "Deep threat analysis"
        timeout_seconds: 600
        validation_required: true
        reflection_enabled: true
        max_iterations: 20

      safe:
        temperature: 0.0
        validation_required: true
        audit_log: true
        description: "Compliance-grade scanning"
        timeout_seconds: 900
        require_approval: true
        output_validation: strict
        pii_detection: true
        compliance_mode: ossa

  # Runtime Configuration
  runtime:
    type: ${AGENT_RUNTIME:-unified}
    supports:
      - google-a2a
      - gitlab-duo
      - ossa-mesh
      - mcp
      - local-execution

    scheduling:
      strategy: ${AGENT_SCHEDULING:-priority}
      priority: ${AGENT_PRIORITY:-high}
      max_concurrent: ${AGENT_MAX_CONCURRENT:-5}
      timeout_seconds: ${AGENT_TIMEOUT:-600}

    resource_limits:
      memory_mb: ${AGENT_MEMORY_MB:-1024}
      cpu_millicores: ${AGENT_CPU_MILLICORES:-1000}

  # Capabilities with Profile Overrides
  capabilities:
    - name: quick_scan
      type: action
      execution_profile: fast
      description: "Quick security triage"
      timeout_seconds: 30

    - name: standard_scan
      type: action
      execution_profile: balanced
      description: "Standard security analysis"
      timeout_seconds: 300

    - name: deep_scan
      type: action
      execution_profile: deep
      description: "Deep threat investigation"
      timeout_seconds: 600

    - name: compliance_scan
      type: action
      execution_profile: safe
      description: "Compliance-grade security audit"
      timeout_seconds: 900

  # Observability
  observability:
    tracing:
      enabled: true
      exporter: otlp
      endpoint: ${OTEL_EXPORTER_OTLP_ENDPOINT}
    metrics:
      enabled: true
      port: 9090
    logging:
      level: ${LOG_LEVEL:-info}
      format: json
      audit_enabled: true

Profile Recommendations

By Use Case

By Industry

Related Specifications

• Memory Model - Agent memory management
• Runtime Specification - Agent lifecycle
• Access Tiers - Permission levels
• LLM Configuration - Provider setup

Version History