Core Architecture
PAI coordinates five specialized AI engines across platforms, enabling seamless multi-model orchestration, local inference, web research, and autonomous code execution.
PNC
Claude Code
Primary DA · Algorithm Runner · ICM Memory
PNK
OpenCode
Multi-File Architecture · TUI Planning
PNG
Gemini CLI
Research · Web Fan-Out · Factual Lookup
PNX
OpenAI Codex
Sandboxed Execution · Adversarial Testing
PNO
Ollama
Local Inference · Zero-Cost Classification
Algorithm v5.7.11
Phase-based development methodology: OBSERVE → THINK → PLAN → BUILD → CHECK → VERIFY
Capabilities Showcase
A comprehensive suite of 129 skills and 33 specialized agents covering research, security, development, and infrastructure management.
🧠 ICM Memory System
Persistent cross-engine memory with semantic search and topic clustering.
- Vector database (Qdrant)
- Entity tracking
- Cross-session state
- Handoff coordination
🔗 Knowledge Graph
Typed entity management with relationship tracking and wikilink traversal.
- People, Companies, Ideas, Research
- 8 relationship types
- Contradiction detection
- Session harvesting
📚 Research Pipeline
Multi-source research automation with AI-powered analysis.
- YouTube + transcript extraction
- arXiv + AI summaries
- NotebookLM integration
- Fabric (240+ patterns)
🔐 Security Capabilities
Enterprise-grade security assessment and threat modeling.
- STRIDE threat modeling
- Vulnerability triage
- Risk assessment workflows
- Penetration testing automation
🛠️ Development Tools
Professional software development workflows with quality enforcement.
- ISA (Ideal State Artifact)
- Phase-based orchestration
- Multi-file architecture
- Quality gates & verification
🤖 Specialized Agents
33 specialized agents for domain-specific tasks.
- CodeReviewer Agency
- SecurityTriage
- QA Testing
- Incident Response
OpenCode Plugins
Extend OpenCode with PAI-native capabilities for cross-platform coordination and advanced workflows.
🔄 Cross-Platform Coordination
Seamless engine-to-engine handoffs and state synchronization.
⚙️ Algorithm Orchestration
Run the PAI Algorithm directly from OpenCode with full phase tracking.
🔬 Research Integration
YouTube, arXiv, and NotebookLM research capabilities as native OpenCode tools.
Infrastructure
Production-grade infrastructure with health monitoring, backup automation, and service orchestration.
📊 Vector Database
Qdrant for semantic search and memory persistence.
8000+ vectors · 384 dimensions · HNSW indexing
🖥️ Local Inference
Ollama on dedicated GPU server for cost-free classification and analysis.
llama3.2:3b · qwen2.5:7b · gemma2:9b · deepseek-r1:7b
⚡ Service Management
systemd orchestration for all PAI services with health probes and automatic recovery.
Example Workflow
See how PAI coordinates multiple engines to deliver comprehensive research and analysis.
# Multi-Engine Research Workflow
1. User: "Research the latest LLM papers on multi-agent systems"
2. PNC (Claude Code) → Orchestrates workflow
└─ Spawns PNG for web research
└─ Spawns PNO for classification
└─ Stores context in ICM
3. PNG (Gemini) → Searches arXiv
└─ Finds 12 relevant papers (cs.AI, cs.MA)
└─ Retrieves AlphaXiv AI summaries
└─ Returns structured data
4. PNO (Ollama) → Classifies papers
└─ Extracts key themes (zero tokens)
└─ Scores relevance
└─ Returns JSON
5. PNC → Synthesizes results
└─ Stores in Knowledge Graph
└─ Creates research note
└─ Surfaces to user
Total Cost: ~4,000 Claude tokens (vs. 20,000+ without PAI)⚠️ Educational Purpose
This infrastructure is built for personal research and educational purposes. Third-party integrations (YouTube, arXiv, NotebookLM, social media platforms) are used in compliance with their respective Terms of Service and API usage policies. All scraping and automation tools are used responsibly and ethically.