Mambu Agentic Platform Overview
The Mambu Agentic Platform is an intelligence layer over your banking data. It lets AI work on that data: gathering it, reasoning over it, and acting on it where permitted, with human oversight. It brings together Mambu's operational systems and an AI layer in a single ecosystem.
What It Does
- From insight to execution. Agents start by retrieving and explaining information, and are being extended to carry out multi-step operations, with a human in the loop where one is configured.
- Built for banking. Agents are designed for banking operations and can be tuned to your own products, policies and exceptions.
- Governed and auditable. Authentication, permissions, guardrails and a full audit trail are built in, so every agent stays within the limits you set and can be reviewed.
Three Layers
The platform is organised in three layers: the agents, the runtime they run on, and the governance that applies to every action.
- Agentic layer. Pre-built, domain-specific agents, from deposits to lending, payments and beyond, with bespoke agents to follow.
- Foundation layer. The runtime and services every agent runs on, including a configurable knowledge base and skills.
- Trust layer. The governance every agent inherits: authentication, permissions, observability, analytics and guardrails.
The Capabilities
The platform comprises four capabilities, which can be used on their own or together.
- Mambu Agents. AI agents that gather data, operate within configurable guardrails, act where they are allowed, and explain every decision.
- Mambu MCP. Governed access to your data, transactions and operations, for agents and any compatible AI client.
- Mambu Insights. A curated data layer designed to support the quality of data used by agents, surfacing actionable signals on growth, risk and anomalies.
- Mambu Process Orchestrator. The engine for multi-step, multi-actor automation, sequencing agents and people with approvals, branching and rollback.
Glossary
Definitions of terms used across the Mambu Agentic Platform.
| Term | Definition |
|---|---|
| Agent | An autonomous AI agent embedded in the Mambu interface that gathers data, reasons against configurable guardrails, acts where it is allowed, and explains every decision. Hosted by Mambu, configured in Agent Studio. |
| MCP | Model Context Protocol, an open industry standard for connecting AI tools to data and systems |
| Mambu MCP | A hosted connection layer that opens Mambu's banking operations to any major MCP-compatible AI client. |
| Ask Mambu | The first agent available on the platform. A read-only chat assistant embedded in the Mambu interface. |
| Agent Studio | The administration interface where agents are defined, configured, and managed. |
| Agent DNA | An agent's identity and purpose, enriched with tenant-specific domain knowledge like products, custom fields, or language preference. |
| Skill | A specific capability of an agent, available to the agent at runtime to process customer input in a predefined manner. |
| Knowledge Base | A collection of information sources an agent can draw from, such as internal policies or user guides. |
| Guardrails | Rules that check an agent's requests and actions before they run, designed to support safety and compliance. |
| Deterministic | A system behaviour that is predictable and repeatable, such as a balance lookup or a guardrail check. |
| Non-deterministic | A system behaviour that varies between runs, such as the wording of an AI-generated answer. |