Mission: Automate the complete broker pipeline from intake to post-close while a multi-agent team continuously improves the product.

• Primary Orchestrator: Tool/MCP routing, policy, inter-agent communications
• TaskGen Orchestrator: Goal→task DAGs, acceptance tests, convergence

• Lead Coder: Primary development and architecture
• Morph/DSPy: Minimal-diff refactors and optimization
• Debug/Aider: Issue resolution and testing
• Small-Code: Specialized small code tasks
• External Reviewer: Code review and quality assurance
• Browser/PC: Web automation and PC interactions
• Voice Agent: Speech recognition and voice UI

\
Project-Nyra/ ├── nyra-core/ # Core orchestration and agents ├── nyra-webapp/ # UI (Open-WebUI + Loab.Chat) ├── nyra-memory/ # Memory systems (memOS + Graphiti + FalkorDB) ├── nyra-infra/ # Infrastructure and deployment └── nyra-prompts/ # Mortgage-specific prompts and workflows \\

1. Start Development Environment: \\�ash docker-compose up -d \\

2. Access Services:

   • NYRA Orchestrator: http://localhost:8000
   • Web UI: http://localhost:3000
   • FalkorDB: localhost:6379
   • ChromaDB: http://localhost:8001

3. Initialize Agents: \\�ash python scripts/initialize-agents.py \\

1. Intake → Customer information capture
2. Pre-Qualification → Initial assessment
3. Pricing → Rate and product selection
4. Documentation → Document collection and processing
5. LOS Integration → Loan origination system
6. Disclosures → Regulatory compliance
7. UW/Appraisal → Underwriting and property valuation
8. Conditions → Condition management
9. Rate Locks → Interest rate securing
10. Clear to Close → Final approval
11. Post-Close → Loan servicing handoff

• memOS (MemoryTensor): Primary memory orchestration
• Graphiti: Relationship and knowledge mapping
• FalkorDB: Graph database for complex relationships
• ChromaDB: Vector storage for semantic search
• KAG-first retrieval: Knowledge-augmented generation

• Open-WebUI: Primary interface
• Loab.Chat: Specialized chat interface
• Xulbux Purple Theme: Consistent branding
• Dify Integration: Planned future enhancement

• Framework: LangGraph + AutoGen2 + OpenEvolve + PraisonAI
• Deployment: Letta for workload-aware deployment
• Observability: Comprehensive monitoring stack
• Memory: Persistent across sessions with learning

• Local: RTX 5090/3090/3060 GPUs
• Cloud: Koyeb + VPS as needed
• Tunnel: Cloudflare Tunnel (ratehunter.net)
• Secrets: Bitwarden → Infisical migration planned

• Mortgage industry compliance built-in
• Audit logging for all operations
• Secure secret management
• Data encryption in transit and at rest

Brand: Xulbux Purple
Mission: Trustworthy, fast, explainable mortgage copilot that learns from every case
Philosophy: Minimal-diff first, ADRs for decisions, humans in the loop