How abdullah00ashraf/managerai works
NexusOrch is a stateful multi-agent orchestration and governance platform. Its documented workflow turns a user goal into an execution plan, resolves conflicts, dispatches domain agents through MCP security checks, then applies risk-based approval or escalation. The graph also preserves the documented RAG, dataset-building, backup, and local observability capabilities. Some links reflect README architecture rather than sampled runtime calls; the implemented Phase-Zero graph in main.py currently wires only the orchestrator, arbitrator, finance, DevOps, and governance nodes.
Main entry points: main.py Graph Runtime · mcp_host.py MCP Tool Host · state.py Execution State · orchestrator_agent.py Goal Planner
Components 24
Orchestration
- Graph Runtime
main.py - Execution State
state.py - Goal Planner
skills/orchestrator_agent.py - Conflict Arbitrator
arbitration/arbitrator_node.py
Agents and Tools
- Finance Agent
skills/finance_agent.py - DevOps Agent
skills/devops_agent.py - Resource Agent
skills/resource_agent.py - Marketing Agent
skills/marketing_agent.py - Communications Agent
skills/comms_agent.py - MCP Tool Host
connector_engine/mcp_host.py
Security and Governance
- Tool Authorization
connector_engine/auth_broker.py - Semantic Guard
connector_engine/semantic_guard.py - Risk Engine
governance/risk_matrix.py
Knowledge and Data
- RAG Population
data_pipeline/populate_rag.py - Vector Store
data_pipeline/vector_store_manager.py - SFT Dataset Builder
data_pipeline/builder/pipeline.py
Operations and Telemetry
- Backup Engine
backup_engine/core.py - Key Vault
backup_engine/vault.py - Local Telemetry Sink
telemetry/local_sqlite_sink.py - Dashboard API
telemetry/dashboard_viewer.py - Telemetry Dashboard
telemetry/dashboard-ui/src/App.jsx
Other components
- User Goal
- External Tools
- Human Review
Connections 24
- User Goal → Graph Runtime: submits goal
README.md - Graph Runtime → Goal Planner: runs first
main.py - Graph Runtime → Execution State: initializes state
main.py - Goal Planner → Execution State: writes plan
skills/orchestrator_agent.py - Graph Runtime → Conflict Arbitrator: runs next
main.py - Conflict Arbitrator → Execution State: resolves conflicts
arbitration/arbitrator_node.py - Graph Runtime → Finance Agent: dispatches
main.py - Graph Runtime → DevOps Agent: dispatches
main.py - Graph Runtime → Risk Engine: gates actions
main.py - Finance Agent → MCP Tool Host: requests tools
README.md - DevOps Agent → MCP Tool Host: requests tools
README.md - Resource Agent → MCP Tool Host: requests tools
README.md - Marketing Agent → MCP Tool Host: requests tools
README.md - Communications Agent → MCP Tool Host: requests tools
README.md - MCP Tool Host → Tool Authorization: authorizes calls
README.md - Tool Authorization → Semantic Guard: checks intent
README.md - Semantic Guard → External Tools: permits execution
README.md - Risk Engine → Human Review: pauses for review
README.md - RAG Population → Vector Store: stores embeddings
README.md - SFT Dataset Builder → RAG Population: provides knowledge
README.md - Backup Engine → Key Vault: gets cipher key
backup_engine/vault.py - Graph Runtime → Local Telemetry Sink: logs spans
main.py - Dashboard API → Local Telemetry Sink: reads spans
telemetry/dashboard_viewer.py - Telemetry Dashboard → Dashboard API: fetches traces
telemetry/dashboard-ui/src/App.jsx