How felipebebas/getnet-multi-agent-support works
Getnet’s product is an authenticated, multi-agent merchant-support service. A client request enters the FastAPI gateway, is checked and routed through a LangGraph workflow, then receives a synthesized and sanitized response; higher-risk cases can pause for operator review and resume. The principal domain stages are knowledge retrieval, merchant operations, and escalation. SQLite-backed history and handoffs, vector retrieval, and observability support the runtime. The README documents capabilities beyond the sampled excerpts; edges without direct evidence are omitted.
Main entry points: graph.py LangGraph Runner · routes.py HTTP Routes · support_tools.py Merchant Tools · escalation_agent.py Escalation Agent
Components 25
API Gateway
- HTTP Routes
src/api/routes.py - Auth and RBAC
src/api/auth.py - Rate Limits
src/api/rate_limit.py - HITL Dashboard
src/api/dashboard.py
Agent Workflow
- LangGraph Runner
src/orchestration/graph.py - Workflow State
src/orchestration/state.py - Intent Router
src/agents/router_agent.py
Support Domains
- Knowledge Agent
src/agents/knowledge_agent.py - Support Agent
src/agents/support_agent.py - Escalation Agent
src/agents/escalation_agent.py - Response Synthesis
src/agents/personality_agent.py - Knowledge Tools
src/tools/knowledge_tools.py - Merchant Tools
src/tools/support_tools.py - Handoff Tools
src/tools/escalation_tools.py
State and Retrieval
- Knowledge Ingestion
src/rag/ingest.py - Vector Retrieval
src/rag/vector_store.py - Session History
src/orchestration/session_store.py - Handoff Persistence
src/orchestration/persistence.py - Merchant Ledger
data/merchants.json
Safety and Operations
- Input Validation
src/guardrails/validator.py - Sensitive Data Masking
src/guardrails/redaction.py - Tracing and Alerts
src/observability/tracer.py - Structured Logging
src/observability/logger.py
Other components
- Merchant / Operator
- Human Operator
Connections 28
- Merchant / Operator → HTTP Routes: sends request
README.md - HTTP Routes → Auth and RBAC: authenticates
src/api/routes.py - HTTP Routes → Rate Limits: applies limits
README.md - HTTP Routes → LangGraph Runner: runs workflow
src/api/routes.py - HTTP Routes → HITL Dashboard: serves dashboard
README.md - LangGraph Runner → Workflow State: uses state
src/orchestration/graph.py - LangGraph Runner → Intent Router: dispatches
src/orchestration/graph.py - LangGraph Runner → Knowledge Agent: dispatches
src/orchestration/graph.py - LangGraph Runner → Support Agent: dispatches
src/orchestration/graph.py - LangGraph Runner → Escalation Agent: dispatches
src/orchestration/graph.py - LangGraph Runner → Response Synthesis: synthesizes with
src/orchestration/graph.py - LangGraph Runner → Input Validation: validates input
src/orchestration/graph.py - LangGraph Runner → Session History: stores turns
src/orchestration/graph.py - LangGraph Runner → Handoff Persistence: records handoffs
src/orchestration/graph.py - Knowledge Agent → Knowledge Tools: uses tools
src/agents/knowledge_agent.py - Knowledge Tools → Vector Retrieval: retrieves chunks
src/tools/knowledge_tools.py - Support Agent → Merchant Tools: uses tools
src/agents/support_agent.py - Merchant Tools → Merchant Ledger: reads merchant data
src/tools/support_tools.py - Merchant Tools → Handoff Persistence: manages tickets
src/tools/support_tools.py - Escalation Agent → Handoff Tools: uses tools
src/agents/escalation_agent.py - Escalation Agent → Merchant Tools: manages tickets
src/agents/escalation_agent.py - Human Operator → HITL Dashboard: reviews handoffs
README.md - HITL Dashboard → HTTP Routes: resumes workflow
README.md - Knowledge Ingestion → Vector Retrieval: indexes knowledge
README.md - LangGraph Runner → Sensitive Data Masking: sanitizes output
README.md - LangGraph Runner → Tracing and Alerts: records execution
src/orchestration/graph.py - Tracing and Alerts → Structured Logging: emits events
src/observability/tracer.py - HTTP Routes → Sensitive Data Masking: sanitizes responses
README.md