> ## 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.

# AI runtime overview

> How Finpace embeds AI into banking safely.

# AI runtime overview

Finpace public positioning presents AI as part of the operating core, not as a chatbot bolted onto the edge.

This documentation therefore models the AI runtime as a **governed execution layer** with five parts:

1. **Agent UI** for customer and staff interaction
2. **MCP gateway** for safe tool exposure
3. **AI Shield** for perimeter and policy enforcement
4. **Policy engine** for action evaluation
5. **AI audit store** for evidence and traceability

## AI design principles

### Assist before acting

AI can summarize, propose, draft, explain, and route before it executes a protected action.

### Tools, not raw system access

AI agents receive task-scoped tools such as:

* retrieve arrangement summary
* draft customer message
* prepare payment instruction
* request covenant waiver review
* summarize application documents

### Policies precede execution

Every state-changing AI call is evaluated for:

* entitlement
* consent
* approval requirement
* product and jurisdiction rule
* risk score
* tool safety profile

### Evidence is mandatory

Each AI-assisted action writes:

* prompt or request context hash
* tool selection
* policy evaluation result
* approvals obtained
* final business outcome