agentOS by Fiserv
Product · API / developer platform by Fiserv
The operating system for financial services agents
agentOS by Fiserv is a governed AI operating system for financial institutions to create, run, and scale AI agents across banking systems.
Category: AI Agents
- Platforms
- Web-based operating system for financial services agents
- Deployment
- Cloud-based with CERT and PROD environments
- Stack Architecture
- 4 Layers
- Deployment Environments
- CERT and PROD
- Agent Types
- Fiserv (1st party) and Vetted Partner (3rd party)
- Governance Features
- Permissioning, Identity, Policy Controls, Audit Logs
- Four-layer governed stack for financial services agents
- Native access to Fiserv cores and data without complex integrations
- Unified credit-based billing across first and third-party agents
- Sandbox (CERT) environment for testing with institution-specific data
Overview
agentOS by Fiserv is an operating system designed for financial services agents, providing a governed platform for banks and financial institutions to deploy agentic AI. It solves the problem of fragmented AI implementation by offering a single four-layer stack that encompasses an agent marketplace, governance, model orchestration, and native data access. It features built-in observability, policy controls, and a unified credit-based billing model. Institutions can test agents in a dedicated certification environment (CERT) using their own data before promoting them to production (PROD). By routing tasks to specialized models or small language models (SLMs), agentOS optimizes both performance and cost for banking workflows.
Core Platform Layers
- Governed AI Stack: The platform provides a single governed stack to create and use agents across Fiserv with built-in observability and controls.
- Agent Marketplace: Users can turn on Fiserv-developed agents and vetted partner agents from a single catalog without complex integration projects.
- Model Orchestration: The system routes each task to the model or specialized Small Language Model (SLM) that performs it best for optimized cost and performance.
- Native Data Access: Agents reach Fiserv data and capabilities directly, enabling native operation across Fiserv products.
Governance and Controls
- Permissioning and Identity: The governance layer includes permissioning and identity management across every agent in the enterprise.
- Policy Controls: Administrators can set policy controls and use kill or suspend functions for any agent.
- Audit and Observability: The platform maintains audit logs and provides observability for all agent activities.
- Optimization and Caching: Built-in caching and optimization help reduce costs and improve performance across different models.
Banking Applications
- Operational Intelligence Systems: Agents help bank teams gain intelligence from legacy systems regardless of their vintage.
- Commercial Loan Onboarding: Specialized agents assist with commercial loan onboarding to drive efficiency.
- Regulatory Monitoring: The system supports regulatory monitoring and controls through automated agent workflows.
Deployment Lifecycle
- Sandbox Testing: Institutions can test any agent with their own data in a certification environment (CERT) before moving to production.
- Self-Service Promotion: Users can certify an agent and promote it to the production environment (PROD) themselves once it passes checks.
- Accuracy Verification: The certification process includes accuracy checks, such as a 99.2% accuracy result shown in platform previews.
- Policy Validation: The platform validates that all policy and controls have passed before allowing an agent to go live.
Use cases
- Bank teams gain operational intelligence from legacy systems regardless of the system's age or vintage.
- Lending departments use agents to automate and streamline the commercial loan onboarding process.
- Compliance officers implement regulatory monitoring and controls through a governed agentic AI stack.
Specifications
- Platforms
- Web-based operating system for financial services agents
- Deployment
- Cloud-based with CERT and PROD environments
- Integrations
- Fiserv cores, Fiserv data, and vetted third-party partner agents
- Stack Architecture
- 4 Layers
- Deployment Environments
- CERT and PROD
- Agent Types
- Fiserv (1st party) and Vetted Partner (3rd party)
- Governance Features
- Permissioning, Identity, Policy Controls, Audit Logs
- Model Support
- LLMs and Specialized SLMs
- Data Connectivity
- Native Fiserv cores and data access
- Billing Model
- Unified Credit System
How it works
- Select Agents: Institutions select first-party Fiserv agents or vetted third-party agents from a central catalog.
- Test in Sandbox
- Verify and Certify: The platform performs accuracy checks and validates that all policy and governance controls are met.
- Promote to Production: Once certified, the user promotes the agent to the production environment (PROD) to run on live data.
Pricing
- Per query
- Pay each time you ask an agent to work
- Per outcome
- Pay when the agent finishes the job
- Subscription
- Flat rate for always-on agents
FAQ
How is agentOS billed?
agentOS uses a unified billing model where institutions buy credits that can be spent across different agents based on their specific value creation.
What are the different credit spending models?
The platform supports three pricing structures: per query (pay each time an agent works), per outcome (pay when the job is finished), and subscription (flat rate for always-on agents).
Can I test agents before paying for them?
Yes, the platform includes a sandbox environment (CERT) where you can test agents with your own institution's data before promoting them to production.
What is the architecture of agentOS?
agentOS is built on a four-layer stack: the Agent Marketplace, a Governance layer, Model orchestration (including SLMs), and a Data layer for native Fiserv product access.
Added by AntSpace from public sources (fiserv.com).