Skip to content

Provisioning And Runtime Topology

LeafEnterprise uses an Azure-first backend topology while preserving the current AWS SQL runtime source where it remains authoritative.

Runtime Components

Component Role
Cloudflare Access / Tunnel Internal perimeter and protected hosted API reachability.
Azure VM / backend process Current hosted backend process for scripts/serve_client_portal.py.
AKS Target heavy execution lane for persistent agent workspaces and worker pools.
Service Bus agent-tasks Durable dispatch queue for agent and long-running backend work.
AWS SQL Server Current claims/reporting source for live savings/audit runtime.
Azure SQL Targeted curated store for Domo takeover, contract/evidence registries, and promoted backend tables.
ADLS Gen2 Bronze/silver evidence, materialized files, artifacts, and replayable source snapshots.
Microsoft Graph / OneDrive Read-only governed contract-packet retrieval.
Azure AI Search / Foundry IQ Indexed evidence and enterprise retrieval context.
APIM Gateway policy layer for AI and future MCP/tool traffic.
Key Vault Runtime secrets, source credentials, and managed identity boundaries.
Azure OpenAI Module copilot, narrative drafting, and agent model calls behind backend guardrails.

Executive Provisioning Summary

Provision Business value
AKS worker lane Gives LeafEnterprise a durable execution plane for long-running agent tasks, data refreshes, validation, and controlled automation.
ADLS Gen2 Keeps evidence and artifacts replayable instead of trapped in dashboard state.
AWS SQL connection Preserves current claims/audit runtime authority while replacement layers mature.
Azure SQL bronze/silver/curated Provides migration and reconciliation structure for Domo takeover without making Domo a runtime dependency.
Microsoft Graph / OneDrive Lets contract packet evidence be retrieved and promoted from governed source files.
Salesforce integration Connects business requests and evidence to backend execution and final deliverables.
Azure AI Search / Foundry IQ Gives agents searchable governed context without exposing raw source systems.
Agent SDKs and workers Allow automation while retaining policy, traces, blockers, and artifact ownership.

Execution Flow

flowchart LR
  UI[Internal UI / dashboard consumer] --> API[LeafEnterprise API]
  API --> SQL[AWS SQL / Azure SQL curated stores]
  API --> Graph[Graph contract packets]
  API --> ADLS[ADLS evidence and artifacts]
  API --> Search[Azure AI Search / Foundry IQ]
  API --> Bus[Service Bus agent-tasks]
  Bus --> AKS[AKS lane workers]
  AKS --> ADLS
  AKS --> SQL
  API --> AOAI[Azure OpenAI via backend guardrails]

Current Versus Target

Area Current posture Target posture
Claims runtime AWS SQL remains active runtime authority. Promote only deliberate governed tables to Azure SQL when ready.
Domo takeover Dedicated Azure SQL/ADLS evidence lane. Complete high-volume extracts, flaw gates, golden comparisons, no runtime Domo dependency.
Agent execution API ledger, local runner, Service Bus dispatch scaffold. AKS persistent lane workspaces with managed identity and artifact replay.
Retrieval Salesforce ITR search-document contract and docs semantic index. Module-split Azure AI Search / Foundry IQ indexes with source-aware routing.
MCP tools Private backend tools and docs-generated route contracts. APIM-governed, policy-aware MCP services generated from machine contracts.

Provisioning Rule

Provisioning docs must separate:

  • what is live now,
  • what is scaffolded and validated locally,
  • what is target architecture,
  • what requires secrets, identity binding, private networking, or operator approval.

This prevents docs from overstating production readiness while still making the intended AKS/ADLS/Azure AI/Search runtime clear to humans and agents.