language-operator 0.0.1 → 0.1.30

data/docs/dsl/constraints.md

@@ -0,0 +1,671 @@
+# Constraints Reference
+
+Complete reference for configuring agent constraints in the Language Operator DSL.
+
+## Table of Contents
+
+- [Overview](#overview)
+- [Time Constraints](#time-constraints)
+- [Resource Constraints](#resource-constraints)
+- [Budget Constraints](#budget-constraints)
+- [Rate Limiting](#rate-limiting)
+- [Content Filtering](#content-filtering)
+- [Complete Examples](#complete-examples)
+
+## Overview
+
+Constraints define limits on agent behavior and resource usage. They help:
+- Prevent runaway costs
+- Enforce rate limits
+- Block inappropriate content
+- Manage resource consumption
+- Ensure timely execution
+
+Define constraints in an agent's `constraints` block:
+
+```ruby
+agent "my-agent" do
+  constraints do
+    timeout '30m'
+    daily_budget 1000
+    requests_per_minute 10
+  end
+end
+```
+
+## Time Constraints
+
+### Timeout
+
+Maximum execution time for the entire agent workflow or autonomous loop iteration.
+
+```ruby
+constraints do
+  timeout '30m'  # 30 minutes
+end
+```
+
+**Format:**
+- `'5m'` - 5 minutes
+- `'2h'` - 2 hours
+- `'30s'` - 30 seconds
+- `'1h30m'` - 1 hour 30 minutes
+
+**Default:** No timeout (runs indefinitely)
+
+**Use cases:**
+- Prevent hung workflows
+- Enforce SLA requirements
+- Manage computational costs
+
+**Example:**
+
+```ruby
+agent "time-sensitive-reporter" do
+  constraints do
+    timeout '15m'  # Must complete within 15 minutes
+  end
+
+  workflow do
+    step :fetch_data do
+      tool 'slow_database_query'
+      timeout '5m'  # Step-level timeout
+    end
+
+    step :analyze do
+      prompt "Analyze: {fetch_data.output}"
+      # Inherits remaining agent timeout
+    end
+  end
+end
+```
+
+### Max Iterations
+
+Maximum number of iterations for autonomous agents or workflow loops.
+
+```ruby
+constraints do
+  max_iterations 100
+end
+```
+
+**Default:** No limit
+
+**Use cases:**
+- Prevent infinite loops
+- Control cost in autonomous mode
+- Enforce bounded execution
+
+**Example:**
+
+```ruby
+agent "bounded-monitor" do
+  mode :autonomous
+
+  constraints do
+    max_iterations 50  # Stop after 50 checks
+  end
+
+  objectives [
+    "Check system status every 5 minutes",
+    "Alert if issues found"
+  ]
+end
+```
+
+## Resource Constraints
+
+### Memory Limit
+
+Maximum memory the agent process can use.
+
+```ruby
+constraints do
+  memory '2Gi'  # 2 Gigabytes
+end
+```
+
+**Format:**
+- `'512Mi'` - 512 Megabytes
+- `'2Gi'` - 2 Gigabytes
+- `'100Mi'` - 100 Megabytes
+
+**Default:** No memory limit (uses container/pod limits)
+
+**Note:** This sets a soft limit. The Kubernetes pod should have matching resource limits.
+
+**Example:**
+
+```ruby
+agent "memory-intensive-processor" do
+  constraints do
+    memory '4Gi'  # Allow up to 4GB
+    timeout '1h'
+  end
+
+  workflow do
+    step :process_large_dataset do
+      tool 'data_processor'
+      # May use significant memory
+    end
+  end
+end
+```
+
+## Budget Constraints
+
+Control costs by limiting LLM API spending.
+
+### Daily Budget
+
+Maximum spending per day (in cents).
+
+```ruby
+constraints do
+  daily_budget 1000  # $10.00 per day
+end
+```
+
+**Units:** Cents (100 = $1.00)
+
+**Behavior:**
+- Agent stops when daily budget exceeded
+- Budget resets at midnight UTC
+- Tracked across all LLM calls
+
+### Hourly Budget
+
+Maximum spending per hour (in cents).
+
+```ruby
+constraints do
+  hourly_budget 100  # $1.00 per hour
+end
+```
+
+**Units:** Cents
+
+**Behavior:**
+- Agent pauses when hourly budget exceeded
+- Resumes at the next hour
+- Useful for rate-limiting expensive operations
+
+### Token Budget
+
+Maximum tokens to consume (input + output combined).
+
+```ruby
+constraints do
+  token_budget 100000  # 100k tokens max
+end
+```
+
+**Units:** Tokens
+
+**Behavior:**
+- Tracks total tokens across all LLM calls
+- Stops when budget exhausted
+- Resets based on agent schedule
+
+### Combined Budget Constraints
+
+```ruby
+agent "cost-controlled-agent" do
+  constraints do
+    daily_budget 2000    # $20/day maximum
+    hourly_budget 200    # $2/hour maximum
+    token_budget 500000  # 500k tokens maximum per day
+  end
+
+  # Agent will stop if ANY limit is reached
+end
+```
+
+**Example:**
+
+```ruby
+agent "budget-conscious-analyst" do
+  description "Analyze data within strict budget limits"
+
+  mode :scheduled
+  schedule "0 9 * * *"  # Daily at 9 AM
+
+  constraints do
+    timeout '30m'
+    daily_budget 500      # $5/day max
+    token_budget 100000   # 100k tokens max
+  end
+
+  workflow do
+    step :analyze do
+      prompt "Provide a concise analysis of today's sales data"
+      # Encouraged to be brief due to token budget
+    end
+  end
+end
+```
+
+## Rate Limiting
+
+Control request frequency to prevent abuse and manage costs.
+
+### Requests Per Minute
+
+```ruby
+constraints do
+  requests_per_minute 10  # Max 10 LLM calls per minute
+end
+```
+
+**Behavior:**
+- Enforced across all LLM API calls
+- Agent pauses if limit exceeded
+- Useful for staying within API quotas
+
+### Requests Per Hour
+
+```ruby
+constraints do
+  requests_per_hour 100  # Max 100 LLM calls per hour
+end
+```
+
+### Requests Per Day
+
+```ruby
+constraints do
+  requests_per_day 1000  # Max 1000 LLM calls per day
+end
+```
+
+### Rate Limit (Generic)
+
+Alternative syntax for any time period:
+
+```ruby
+constraints do
+  rate_limit requests: 60, period: '1h'  # 60 requests per hour
+end
+```
+
+### Combined Rate Limits
+
+```ruby
+agent "rate-limited-agent" do
+  constraints do
+    requests_per_minute 5    # Burst protection
+    requests_per_hour 200    # Hourly cap
+    requests_per_day 2000    # Daily cap
+  end
+
+  # Agent respects all limits (most restrictive applies)
+end
+```
+
+**Example:**
+
+```ruby
+agent "monitoring-agent" do
+  mode :autonomous
+
+  constraints do
+    requests_per_minute 2   # Check every 30 seconds max
+    requests_per_day 2880   # 2 per minute * 60 min * 24 hr
+    daily_budget 500        # $5/day budget
+  end
+
+  objectives [
+    "Monitor system health continuously",
+    "Alert on anomalies"
+  ]
+end
+```
+
+## Content Filtering
+
+Block or filter inappropriate content and topics.
+
+### Blocked Topics
+
+Prevent the agent from processing certain topics:
+
+```ruby
+constraints do
+  blocked_topics ['violence', 'illegal-activity', 'adult-content']
+end
+```
+
+**Behavior:**
+- Agent rejects requests containing blocked topics
+- Content moderation applied to inputs and outputs
+- Logged for audit purposes
+
+### Blocked Patterns
+
+Block content matching regex patterns:
+
+```ruby
+constraints do
+  blocked_patterns [
+    /\b\d{3}-\d{2}-\d{4}\b/,  # SSN pattern
+    /\b\d{16}\b/,              # Credit card number
+    /password\s*[:=]/i         # Password disclosure
+  ]
+end
+```
+
+**Behavior:**
+- Regex matched against all inputs and outputs
+- Request rejected if pattern found
+- Helps prevent data leakage
+
+### Combined Content Filtering
+
+```ruby
+agent "safe-agent" do
+  constraints do
+    blocked_topics [
+      'violence',
+      'hate-speech',
+      'illegal-activity',
+      'self-harm'
+    ]
+
+    blocked_patterns [
+      /\b\d{3}-\d{2}-\d{4}\b/,           # SSN
+      /\b\d{16}\b/,                       # Credit card
+      /\b[A-Z0-9._%+-]+@[A-Z0-9.-]+\.[A-Z]{2,}\b/i  # Email addresses
+    ]
+  end
+end
+```
+
+**Example:**
+
+```ruby
+agent "customer-service-agent" do
+  description "Safe customer service assistant"
+
+  mode :reactive
+
+  persona <<~PERSONA
+    You are a helpful customer service representative.
+    Always maintain a professional and respectful tone.
+  PERSONA
+
+  constraints do
+    # Safety constraints
+    blocked_topics [
+      'violence',
+      'hate-speech',
+      'illegal-activity'
+    ]
+
+    # PII protection
+    blocked_patterns [
+      /\b\d{3}-\d{2}-\d{4}\b/,  # SSN
+      /\b\d{16}\b/,              # Credit card
+    ]
+
+    # Rate limiting
+    requests_per_minute 30
+    requests_per_hour 500
+
+    # Budget
+    hourly_budget 100  # $1/hour
+    daily_budget 1000  # $10/day
+
+    # Time limits
+    timeout '5m'
+  end
+end
+```
+
+## Complete Examples
+
+### Production ETL Agent
+
+```ruby
+agent "production-etl" do
+  description "Production-grade ETL with comprehensive constraints"
+
+  mode :scheduled
+  schedule "0 2 * * *"  # 2 AM daily
+
+  constraints do
+    # Time constraints
+    timeout '2h'           # Must complete within 2 hours
+    max_iterations 10      # Max 10 retry attempts
+
+    # Resource constraints
+    memory '8Gi'          # Allow 8GB memory for large datasets
+
+    # Budget constraints
+    daily_budget 2000     # $20/day maximum
+    token_budget 1000000  # 1M tokens max
+
+    # Rate limiting (avoid API throttling)
+    requests_per_minute 30
+    requests_per_hour 1000
+
+    # Content filtering
+    blocked_patterns [
+      /\b\d{3}-\d{2}-\d{4}\b/,  # Prevent SSN in logs
+      /password\s*[:=]/i         # Prevent password leakage
+    ]
+  end
+
+  workflow do
+    step :extract do
+      tool 'database_extract'
+      timeout '30m'  # Step-level timeout
+    end
+
+    step :transform do
+      depends_on :extract
+      tool 'data_transform'
+      timeout '45m'
+    end
+
+    step :load do
+      depends_on :transform
+      tool 'warehouse_load'
+      timeout '30m'
+      retry_on_failure max_attempts: 3
+    end
+  end
+end
+```
+
+### Cost-Optimized Analysis Agent
+
+```ruby
+agent "frugal-analyst" do
+  description "Cost-optimized analysis with strict budgets"
+
+  mode :scheduled
+  schedule "0 9 * * 1-5"  # Weekdays at 9 AM
+
+  constraints do
+    # Aggressive budget controls
+    daily_budget 100        # Only $1/day
+    hourly_budget 50        # $0.50/hour
+    token_budget 50000      # 50k tokens max
+
+    # Rate limits to spread requests
+    requests_per_minute 2   # Slow and steady
+    requests_per_hour 60
+
+    # Reasonable timeout
+    timeout '30m'
+  end
+
+  workflow do
+    step :analyze do
+      prompt "Provide a BRIEF analysis (under 200 words) of today's key metrics"
+      # Prompt emphasizes brevity to save tokens
+    end
+
+    step :distribute do
+      depends_on :analyze
+      tool 'send_email'
+      params(
+        to: 'team@company.com',
+        subject: 'Daily Brief',
+        body: '{analyze.output}'
+      )
+    end
+  end
+end
+```
+
+### High-Throughput Webhook Handler
+
+```ruby
+agent "webhook-processor" do
+  description "Handle high-volume webhook events"
+
+  mode :reactive
+
+  webhook "/api/events" do
+    method :post
+    authenticate { verify_api_key }
+  end
+
+  constraints do
+    # High throughput settings
+    requests_per_minute 120   # 2/second burst
+    requests_per_hour 5000    # 83/minute sustained
+    requests_per_day 50000    # Room for spikes
+
+    # Generous budget for high volume
+    hourly_budget 500    # $5/hour
+    daily_budget 10000   # $100/day
+
+    # Quick timeout per event
+    timeout '30s'  # Process each event fast
+
+    # Memory for concurrent processing
+    memory '4Gi'
+
+    # Safety
+    blocked_topics ['spam', 'malicious-content']
+  end
+
+  on_webhook_event do |event|
+    # Process event quickly within 30s timeout
+  end
+end
+```
+
+### Secure Customer Data Agent
+
+```ruby
+agent "secure-data-processor" do
+  description "Process customer data with strong PII protection"
+
+  mode :scheduled
+  schedule "0 0 * * *"  # Midnight daily
+
+  constraints do
+    # Time and resource limits
+    timeout '1h'
+    memory '4Gi'
+
+    # Budget
+    daily_budget 1000  # $10/day
+
+    # Comprehensive PII filtering
+    blocked_patterns [
+      # SSN
+      /\b\d{3}-\d{2}-\d{4}\b/,
+      # Credit card
+      /\b(?:\d{4}[-\s]?){3}\d{4}\b/,
+      # Phone numbers
+      /\b\d{3}[-.]?\d{3}[-.]?\d{4}\b/,
+      # Email addresses
+      /\b[A-Z0-9._%+-]+@[A-Z0-9.-]+\.[A-Z]{2,}\b/i,
+      # IP addresses
+      /\b(?:\d{1,3}\.){3}\d{1,3}\b/,
+      # API keys (common formats)
+      /\b[A-Za-z0-9_-]{32,}\b/
+    ]
+
+    # Sensitive topics
+    blocked_topics [
+      'medical-records',
+      'financial-details',
+      'authentication-credentials'
+    ]
+
+    # Rate limiting
+    requests_per_minute 10
+  end
+
+  workflow do
+    step :process_data do
+      tool 'data_anonymizer'
+      # Ensures PII is removed before analysis
+    end
+
+    step :analyze_anonymized do
+      depends_on :process_data
+      prompt "Analyze this anonymized data: {process_data.output}"
+    end
+  end
+end
+```
+
+## Best Practices
+
+### Time Constraints
+
+1. **Always set timeouts for production** - Prevent indefinite hangs
+2. **Set realistic limits** - Allow enough time for completion
+3. **Use step-level timeouts** - Granular control over slow operations
+4. **Monitor timeout events** - Optimize workflows that frequently timeout
+
+### Resource Constraints
+
+1. **Match Kubernetes limits** - Ensure container resources align
+2. **Test with realistic data** - Verify memory limits are sufficient
+3. **Monitor resource usage** - Track actual consumption vs. limits
+
+### Budget Constraints
+
+1. **Start conservative** - Begin with low budgets, increase as needed
+2. **Use hourly + daily limits** - Prevent burst spending
+3. **Track costs** - Monitor actual spending vs. budgets
+4. **Set token budgets** - Prevent excessive LLM usage
+5. **Review monthly** - Adjust budgets based on actual costs
+
+### Rate Limiting
+
+1. **Layer limits** - Use minute, hour, and day limits together
+2. **Account for bursts** - Minute limit should allow reasonable bursts
+3. **Match API quotas** - Stay within provider rate limits
+4. **Monitor rejections** - Track how often limits are hit
+
+### Content Filtering
+
+1. **Block sensitive patterns** - Always filter PII (SSN, credit cards, etc.)
+2. **Use broad patterns** - Catch variations (with/without dashes, etc.)
+3. **Test filters** - Verify patterns match intended content
+4. **Log blocks** - Audit what content is being filtered
+5. **Review regularly** - Update patterns as new risks emerge
+
+### Combined Constraints
+
+1. **Use multiple constraint types** - Defense in depth
+2. **Make budgets compatible** - Ensure daily budget allows hourly * 24
+3. **Test limits** - Verify constraints work as expected
+4. **Document rationale** - Explain why specific limits were chosen
+5. **Monitor violations** - Track which constraints are hit most often
+
+## See Also
+
+- [Agent Reference](agent-reference.md) - Complete agent DSL reference
+- [Workflows](workflows.md) - Workflow definition guide
+- [Best Practices](best-practices.md) - Production deployment patterns
+- [Webhooks](webhooks.md) - Reactive agent configuration