> ## Documentation Index
> Fetch the complete documentation index at: https://docs.finpace.tech/llms.txt
> Use this file to discover all available pages before exploring further.

# Platform architecture

> Reference architecture for Finpace core banking, channels, integrations, and governed AI runtime.

# Platform architecture

## Logical layers

```text theme={null}
Channels and workbenches
  ├─ Retail and business digital channels
  ├─ RM and operations workbench
  ├─ Contact-center and assisted-service interfaces
  ├─ Partner channels
  └─ Agent UI

AI runtime and control plane
  ├─ MCP gateway
  ├─ Policy engine
  ├─ Tool registry
  ├─ AI Shield
  └─ AI audit and evidence store

Core banking domain services
  ├─ Party and customer
  ├─ Agreements and entitlements
  ├─ Product factory, pricing, campaigns, experiments
  ├─ Arrangements and accounts
  ├─ Posting, balances, holds, and liquidity
  ├─ Lending origination, affordability, scoring, servicing
  ├─ Collateral, covenants, and conditions
  ├─ Cases, documents, tasks, notifications
  └─ Interaction history and workbench context

Shared control services
  ├─ Identity and access
  ├─ Approval orchestration
  ├─ Audit and immutable event log
  ├─ Reporting and reconciliation
  └─ Event streaming

External integrations
  ├─ KYC and KYB
  ├─ Sanctions and AML
  ├─ Payment rails
  ├─ Document signing
  ├─ Credit bureaus and open banking
  └─ CRM, DWH, and analytics
```

## Architectural rules

### 1. Core APIs remain authoritative

Posting authority, arrangement state transitions, credit decisions, limit enforcement, and servicing outcomes are executed by core services.

### 2. Agents use governed tools

Finpace MCP exposes business-safe tools rather than unrestricted CRUD access to core services.

### 3. Protected actions are policy-evaluated

Before execution, the runtime evaluates:

* actor identity
* delegated authority
* channel
* customer consent
* product and jurisdiction rules
* approval profile
* policy outcome
* risk score

### 4. Evidence objects are reusable

The same evidence objects support:

* internal audit
* regulator review
* customer dispute handling
* operational traceability
* AI explainability

## Deployment model

The reference architecture supports:

* private cloud
* client cloud
* hybrid deployment
* controlled on-premises deployment for high-sensitivity institutions

Regional isolation is available for:

* prompts
* retrieval stores
* tool execution logs
* document caches
* policy packs
* audit evidence