How we built an AI agent for field associates with Red Hat AI

Sellers need information in order to do their jobs. At most organizations, this information already exists in various internal systems. The challenge facing every sales team isn't finding that information; it's turning it into action fast enough to matter. 

To address this challenge, we built Sales Assistant, an enterprise AI agent powered by Red Hat AI and used by Red Hat sellers internally. Sales Assistant surfaces appropriate context, takes action, completes tasks, streamlines workflows, and frees sellers to focus on customers.

For Red Hat sellers, critical data lives across Salesforce, product documentation, pricing systems, lifecycle databases, and internal knowledge bases. Each system serves an important purpose, but together they can create fragmented workflows. Preparing for customer conversations often means switching between multiple applications to gather context, validate information, and complete routine tasks.

Red Hat sellers didn’t lack data or tooling—but navigating disconnected systems produced operational friction.

From questions to action in seconds

With Sales Assistant, sellers can interact with enterprise systems using natural language prompts. A seller preparing for a renewal can ask, "What's the renewal strategy for this account?" and receive a grounded, cited response in seconds. Another seller can request, "Generate a quote for opportunity XYZ," and the agent prepares it for human review before submission.

Instead of searching across multiple applications, sellers interact through a single interface that connects to the enterprise systems where their data already resides.

Behind the scenes, Sales Assistant orchestrates specialized sub-agents that interact with Salesforce, customer engagement platforms, pricing and configuration systems, lifecycle databases, and internal knowledge repositories. Every response is grounded in enterprise data, giving sellers visibility into the sources behind each recommendation. 

Sales Assistant also maintains conversational context across multiple interactions. Sellers can ask follow-up questions like "What products is this customer currently using?" without repeating previous context, enabling a more natural and efficient workflow.

Sales Assistant fits into existing workflows. Sellers can interact with it within the tools they already use, including Salesforce, Slack, a web interface, and a mobile application.

Running in production, not in a lab

Sales Assistant is not a prototype or proof of concept. It is a production AI agent processing thousands of requests every day.

Running AI at production scale requires more than deploying an LLM. Sales Assistant is built on Red Hat AI using a layered architecture (see Figure 1) designed to deliver scalability, governance, security, and operational reliability. 

Figure 1. Sales Assistant architecture diagram showing orchestration, model inference, deployment, autoscaling, and observability

Users authenticate through Red Hat’s single sign-on technology, which is based on Keycloak. After authentication, requests are routed to the Supervisor Agent running on Red Hat OpenShift AI.

The Supervisor Agent orchestrates specialized agents, manages tool execution through Model Context Protocol (MCP) servers registered in the Red Hat OpenShift AI MCP registry, and coordinates access to enterprise systems with centralized governance and observability.

Red Hat AI Inference handles model inference using a hybrid strategy that automatically selects the most suitable hosted or frontier models for each request, balancing latency, performance, and cost.

To ground responses in enterprise knowledge, the platform continuously ingests documents—more than 300,000 so far—using data pipelines built with KubeRay and Docling on Red Hat OpenShift AI. This retrieval-augmented generation (RAG) pipeline grounds responses in current enterprise knowledge. At query time, the platform retrieves the most relevant information, re-ranks results using models served through Red Hat AI Inference, and injects only the highest-value context into the model prompt.

The application platform is standardized on Red Hat Universal Base Image based on Red Hat Enterprise Linux, while backend services are built with Quarkus to provide lightweight, cloud-native Java services optimized for Kubernetes.

Application promotion across development, staging, and production environments is managed through Kustomize. Horizontal Pod Autoscalers dynamically adjust capacity as demand changes, while observability provides operational visibility across the entire platform. 

Helping sellers focus on customers

By bringing enterprise data and business actions together in a single conversational experience, Sales Assistant reduces the time sellers spend searching for information and navigating disconnected systems.

Sales Assistant isn’t yet another tool or system sellers need to navigate; it’s an intelligent interface for existing enterprise tools, and helps sellers prepare faster, respond with greater confidence, and spend more time focused on customer conversations.

For Red Hat, Sales Assistant demonstrates that enterprise AI is most valuable when it integrates directly into existing workflows while meeting the governance, security, and operational standards required for production.

Looking ahead

Sales Assistant continues to evolve as Red Hat expands its enterprise AI platform.

The next steps in its development will focus on transitioning to a skill-based architecture, expanding support for the Red Hat Services and Ecosystem portfolios, improving enterprise knowledge management, deepening user personalization, and broadening agent capabilities to enable sellers to complete more tasks directly through the assistant. The platform is also evolving toward more proactive and adaptive AI agents that can anticipate user needs while continuing to operate within enterprise governance, security, and compliance boundaries.

Building enterprise AI for production

Enterprises no longer have to prove that AI models are capable of doing real work. Their next challenge is to deploy AI tools that securely connect to enterprise data, integrate with business workflows, and operate reliably at scale. 

Built on Red Hat AI, Sales Assistant combines orchestration, retrieval, governance, and observability into a production-ready platform for enterprise deployments. 

As organizations move beyond experimentation, success depends on more than model performance. It requires a platform that provides consistency across hybrid environments, integrates with existing enterprise systems, and delivers the operational controls needed for production deployments. Sales Assistant demonstrates what that looks like in practice.

Explore Red Hat AI to learn more.