orka-reasoning 0.7.1

Modular agent orchestrator for reasoning pipelines

OrKa-Reasoning

AI Orchestration with 100x Faster Vector Search - OrKa transforms your AI workflows with YAML-driven agent orchestration, intelligent memory management, and lightning-fast semantic search powered by RedisStack HNSW indexing.

---

🚀 What's New in V0.7.0

• 🚀 100x Faster Vector Search - RedisStack HNSW indexing now default across all components
• ⚡ Sub-millisecond Search Latency - O(log n) complexity for massive datasets
• 🏗️ Unified Architecture - All components now use RedisStack with intelligent fallback
• 🖥️ Professional CLI Dashboard - Real-time performance monitoring and metrics
• 🔧 Zero-Breaking Migration - Complete backward compatibility maintained

---

⚡ 2-Minute Quickstart

Get OrKa running with enterprise-grade performance in 2 minutes:

Prerequisites: Ensure Docker is installed and running on your system.

# 1. Install OrKa with all dependencies
pip install orka-reasoning fastapi uvicorn kafka-python

# 2. Set your OpenAI key
export OPENAI_API_KEY=your-key-here

# 3. Start OrKa (automatically includes RedisStack + 100x faster vector search)
# For LOCAL development:
python -m orka.orka_start

# For PRODUCTION with Kafka streaming:
python -m orka.start_kafka

In a separate terminal:

# 4. Create a simple workflow
cat > quickstart.yml << EOF
orchestrator:
  id: intelligent-qa
  strategy: sequential
  agents: [classifier, memory_search, web_search, answer_builder, memory_store]

agents:
  - id: classifier
    type: openai-classification
    prompt: "Classify this query type"
    options: [factual_question, how_to_guide, current_events, opinion]

  - id: memory_search
    type: memory-reader
    namespace: knowledge_base
    params:
      limit: 5
      enable_context_search: true
      similarity_threshold: 0.8
    prompt: "Find relevant information about: {{ input }}"

  - id: web_search
    type: duckduckgo
    prompt: "{{ input }}"

  - id: answer_builder
    type: openai-answer
    prompt: |
      Query: {{ input }}
      Type: {{ previous_outputs.classifier }}
      Memory: {{ previous_outputs.memory_search }}
      Web Results: {{ previous_outputs.web_search }}
      
      Build a comprehensive answer combining memory and web results.

  - id: memory_store
    type: memory-writer
    namespace: knowledge_base
    params:
      vector: true
      metadata:
        query_type: "{{ previous_outputs.classifier }}"
        has_web_results: "{{ previous_outputs.web_search | length > 0 }}"
        confidence: "high"
    prompt: |
      Query: {{ input }}
      Answer: {{ previous_outputs.answer_builder }}
      Sources: Web + Memory
EOF

# 5. Run your intelligent AI workflow
python -m orka.orka_cli ./quickstart.yml "What are the latest developments in quantum computing?"

# 6. Monitor performance (in another terminal)
python -m orka.orka_cli memory watch

# 7. Optional: Run OrKa UI for visual workflow monitoring
docker pull marcosomma/orka-ui:latest
docker run -it -p 80:80 --name orka-ui marcosomma/orka-ui:latest
# Then open http://localhost in your browser

This creates an intelligent Q&A system that:

• Classifies your query type
• Searches existing knowledge with 100x faster vector search
• Gets fresh information from the web
• Combines both sources into a comprehensive answer
• Stores the interaction for future reference

---

Performance Comparison

Setup	Vector Search	Typical Use Case	Performance
RedisStack	✅ HNSW Indexing	Production AI, semantic search	100x faster
Basic Redis	❌ Text only	Development, simple workflows	Standard

---

🧠 Hands-On: Build Your First AI Memory System

Let's build a conversational AI that remembers and learns from interactions:

Step 1: Create the Workflow

# conversational-ai.yml
orchestrator:
  id: conversational-ai
  strategy: sequential
  memory_config:
    decay:
      enabled: true
      default_short_term_hours: 2    # Conversations fade after 2 hours
      default_long_term_hours: 168   # Important info lasts 1 week
  agents:
    - conversation_context
    - interaction_classifier  
    - response_generator
    - memory_storage

agents:
  # Retrieve relevant conversation history
  - id: conversation_context
    type: memory-reader
    namespace: user_conversations
    params:
      limit: 5
      enable_context_search: true
      context_weight: 0.4
      temporal_weight: 0.3
      enable_temporal_ranking: true
    prompt: "Find relevant conversation history for: {{ input }}"

  # Understand the interaction type
  - id: interaction_classifier
    type: openai-classification
    prompt: |
      Based on history: {{ previous_outputs.conversation_context }}
      Current input: {{ input }}
      
      Classify this interaction:
    options: [question, followup, correction, new_topic, feedback]

  # Generate contextually aware response
  - id: response_generator
    type: openai-answer
    prompt: |
      Conversation history: {{ previous_outputs.conversation_context }}
      Interaction type: {{ previous_outputs.interaction_classifier }}
      Current input: {{ input }}
      
      Generate a response that acknowledges the conversation history
      and maintains continuity.

  # Store the interaction with intelligent classification
  - id: memory_storage
    type: memory-writer
    namespace: user_conversations
    params:
      vector: true  # Enable semantic search
      metadata:
        interaction_type: "{{ previous_outputs.interaction_classifier }}"
        has_history: "{{ previous_outputs.conversation_context | length > 0 }}"
        response_quality: "pending_feedback"
        timestamp: "{{ now() }}"
    # memory_type automatically classified based on importance
    prompt: |
      User: {{ input }}
      Type: {{ previous_outputs.interaction_classifier }}
      Assistant: {{ previous_outputs.response_generator }}

Step 2: Run and Monitor

# Start the conversation
python -m orka.orka_cli ./conversational-ai.yml "Hello, I'm working on a machine learning project"

# Continue the conversation (it will remember context)
python -m orka.orka_cli ./conversational-ai.yml "What algorithms would you recommend for image classification?"

# Monitor memory performance in real-time
python -m orka.orka_cli memory watch

You'll see a professional dashboard like this:

┌─────────────────────────────────────────────────────────────┐
│ OrKa Memory Dashboard - 14:23:45 | Backend: redisstack     │
├─────────────────────────────────────────────────────────────┤
│ 🔧 Backend: redisstack (HNSW)  ⚡ Decay: ✅ Enabled        │
│ 📊 Memories: 1,247            📝 Active: 1,224             │
│ 🚀 HNSW Performance: 1,203     Avg: 2.1ms | Hybrid: 856   │
│ 🧠 Memory Types: Short: 423    💾 Long: 801 | 🔥 Recent   │
└─────────────────────────────────────────────────────────────┘

Step 3: Verify Memory Learning

# Check what the AI remembers about you
python -m orka.orka_cli memory stats

# Search specific memories
redis-cli FT.SEARCH orka:mem:idx "@namespace:user_conversations machine learning" LIMIT 0 5

---

📚 Ready-to-Use Workflow Templates

1. Intelligent Q&A with Web Search

# intelligent-qa.yml
orchestrator:
  id: smart-qa
  strategy: sequential
  agents: [search_needed, router, web_search, answer_with_sources]

agents:
  - id: search_needed
    type: openai-binary
    prompt: "Does this question require recent information? {{ input }}"

  - id: router
    type: router
    params:
      decision_key: search_needed
      routing_map:
        "true": [web_search, answer_with_sources]
        "false": [answer_with_sources]

  - id: web_search
    type: duckduckgo
    prompt: "{{ input }}"

  - id: answer_with_sources
    type: openai-answer
    prompt: |
      Question: {{ input }}
      {% if previous_outputs.web_search %}
      Web Results: {{ previous_outputs.web_search }}
      {% endif %}
      
      Provide a comprehensive answer with sources.

2. Content Analysis Pipeline

# content-analyzer.yml
orchestrator:
  id: content-analysis
  strategy: parallel
  agents: [fork_analysis, join_results]

agents:
  - id: fork_analysis
    type: fork
    targets:
      - [sentiment_analysis]
      - [topic_classification]
      - [toxicity_check]

  - id: sentiment_analysis
    type: openai-classification
    options: [positive, negative, neutral]
    prompt: "Analyze sentiment: {{ input }}"

  - id: topic_classification
    type: openai-classification
    options: [tech, business, science, politics, sports]
    prompt: "Classify topic: {{ input }}"

  - id: toxicity_check
    type: openai-binary
    prompt: "Is this content toxic or inappropriate? {{ input }}"

  - id: join_results
    type: join
    prompt: |
      Combine analysis results:
      Sentiment: {{ previous_outputs.sentiment_analysis }}
      Topic: {{ previous_outputs.topic_classification }}
      Safe: {{ previous_outputs.toxicity_check }}

3. Knowledge Base Builder

# knowledge-builder.yml
orchestrator:
  id: knowledge-builder
  strategy: sequential
  memory_config:
    decay:
      enabled: true
      default_long_term_hours: 720  # Keep knowledge for 30 days
  agents: [fact_checker, knowledge_storer, knowledge_retriever]

agents:
  - id: fact_checker
    type: openai-answer
    prompt: |
      Verify this information and rate confidence (1-10):
      {{ input }}

  - id: knowledge_storer
    type: memory-writer
    namespace: knowledge_base
    params:
      memory_type: long_term
      vector: true
      metadata:
        confidence: "{{ previous_outputs.fact_checker.confidence | default('unknown') }}"
        verified: "{{ previous_outputs.fact_checker.verified | default(false) }}"
        domain: "{{ previous_outputs.fact_checker.domain | default('general') }}"
    prompt: |
      Fact: {{ input }}
      Verification: {{ previous_outputs.fact_checker }}

  - id: knowledge_retriever
    type: memory-reader
    namespace: knowledge_base
    params:
      limit: 10
      similarity_threshold: 0.8
    prompt: "Find related knowledge: {{ input }}"

---

🎯 Agent Quick Reference

Memory Agents (Powered by RedisStack HNSW)

# Read memories with 100x faster search
- id: memory_search
  type: memory-reader
  namespace: my_namespace
  params:
    limit: 10                        # Max results
    enable_context_search: true      # Use conversation context
    similarity_threshold: 0.8        # Relevance threshold
    enable_temporal_ranking: true    # Boost recent memories
  prompt: "Search for: {{ input }}"

# Store memories with intelligent decay
- id: memory_store
  type: memory-writer
  namespace: my_namespace
  params:
    vector: true                     # Enable semantic search
    # memory_type: auto-classified   # short_term or long_term
    metadata:
      source: "user_input"
      confidence: "high"
      timestamp: "{{ now() }}"
  prompt: "Store: {{ input }}"

LLM Agents

# Binary classification
- id: yes_no_classifier
  type: openai-binary
  prompt: "Is this a question? {{ input }}"

# Multi-class classification  
- id: topic_classifier
  type: openai-classification
  options: [tech, science, business, other]
  prompt: "Classify: {{ input }}"

# Answer generation
- id: answer_builder
  type: openai-answer
  prompt: |
    Context: {{ previous_outputs.context }}
    Question: {{ input }}
    Generate a detailed answer.

Routing & Control Flow

# Dynamic routing
- id: content_router
  type: router
  params:
    decision_key: content_type
    routing_map:
      "question": [search_agent, answer_agent]
      "statement": [fact_checker]

# Parallel processing
- id: parallel_validator
  type: fork
  targets:
    - [sentiment_check]
    - [toxicity_check]
    - [fact_validation]

# Wait for parallel completion
- id: combine_results
  type: join
  prompt: "Combine all validation results"

---

🖥️ CLI Commands You'll Use Daily

# Real-time memory monitoring with RedisStack metrics
orka memory watch --interval 3

# Check memory statistics and performance
orka memory stats

# Clean up expired memories (with HNSW index optimization)
orka memory cleanup --dry-run

# View configuration and backend status
orka memory configure

# Run workflows
orka run ./my-workflow.yml "Your input here"

# Check system health
orka system status

---

🚀 Production Deployment

Docker Compose Setup

# docker-compose.yml
version: '3.8'
services:
  redis-stack:
    image: redis/redis-stack:latest
    ports:
      - "6379:6379"
    volumes:
      - redis_data:/data
    environment:
      - REDIS_ARGS=--save 60 1000

  orka:
    build: .
    ports:
      - "8000:8000"
    environment:
      - REDIS_URL=redis://redis-stack:6379/0
      - ORKA_MEMORY_BACKEND=redisstack
      - OPENAI_API_KEY=${OPENAI_API_KEY}
    depends_on:
      - redis-stack

volumes:
  redis_data:

Environment Variables

# Core settings
export OPENAI_API_KEY=your-key-here
export ORKA_MEMORY_BACKEND=redisstack
export REDIS_URL=redis://localhost:6379/0

# Performance tuning
export ORKA_MAX_CONCURRENT_REQUESTS=100
export ORKA_TIMEOUT_SECONDS=300

# Memory management
export ORKA_MEMORY_DECAY_ENABLED=true
export ORKA_DEFAULT_SHORT_TERM_HOURS=2
export ORKA_DEFAULT_LONG_TERM_HOURS=168

---

🔧 Migration from Basic Redis

Upgrade to RedisStack for 100x performance improvement:

# 1. Analyze your current memories
python scripts/migrate_to_redisstack.py --dry-run

# 2. Backup existing data
redis-cli BGSAVE

# 3. Start RedisStack
docker run -d -p 6379:6379 --name orka-redis redis/redis-stack:latest

# 4. Migrate your memories
python scripts/migrate_to_redisstack.py --migrate

# 5. Validate migration
python scripts/migrate_to_redisstack.py --validate

# 6. Update your applications (no code changes needed!)
export ORKA_MEMORY_BACKEND=redisstack

---

🐛 Troubleshooting

Common Issues & Quick Fixes

Issue	Quick Fix
"unknown command 'FT.CREATE'"	You're using basic Redis. Install RedisStack: docker run -d -p 6379:6379 redis/redis-stack:latest
"Cannot connect to Redis"	Check Redis is running: redis-cli ping
Memory search returns no results	Check vector indexing: redis-cli FT._LIST
Slow performance	Verify RedisStack HNSW: orka memory configure
Out of memory errors	Run cleanup: orka memory cleanup

Performance Optimization

# Check current performance
orka memory watch

# Optimize HNSW index
redis-cli FT.CONFIG SET FORK_GC_CLEAN_THRESHOLD 100

# Monitor Redis memory usage
redis-cli INFO memory

---

📊 Performance Benchmarks

Metric	Basic Redis	RedisStack HNSW	Improvement
Vector Search	50-200ms	0.5-5ms	100x faster
Memory Usage	100% baseline	40%	60% reduction
Throughput	1,000/sec	50,000/sec	50x higher
Concurrent Searches	10-50	1,000+	20x more

---

📚 Learn More

• 🎥 Video Tutorial - 5-minute OrKa overview
• 📚 Full Documentation - Complete API reference
• 💬 Community Discord - Get help and share workflows
• 🐛 GitHub Issues - Report bugs and request features

---

🤝 Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

📜 License

Apache 2.0 License. See LICENSE for details.

---

Ready to supercharge your AI workflows?

pip install orka-reasoning && docker run -d -p 6379:6379 redis/redis-stack:latest

Get Started Now →