Global technology company Syntegon puts 500+ RAG agents to work with Open WebUI

Overview #

Syntegon built Syntia (Syntegon Intelligence Assistance) on Open WebUI and hosts it on AWS. It gives employees across engineering, operations, compliance and other functions one SSO-protected, fully auditable platform for 8+ AI models and RAG agents.

At a Glance #

• Industry: Global technology company
• Company size: 7,300 employees at 49 locations in more than 20 countries; headquartered in Stuttgart, Germany
• Go-live: December 2025
• Active users: 2,000–2,500 in the last 14 days, at the time of reporting
• Access: Licensed for all Syntegon employees globally, except in regions where compliance requirements and local regulations restrict access
• Deployment: Amazon EKS on AWS Fargate in the EU, with Kubernetes autoscaling; EU-only data residency
• Models: 8+ models across Azure OpenAI and AWS Bedrock, including OpenAI GPT and Anthropic Claude; GPT-5-mini is the default
• Integrations: SharePoint, Confluence, Jira and web search
• Key results: 500+ active RAG agents; Syntia is the only AI platform at Syntegon approved under its data governance and Work Council framework

About Syntegon #

Syntegon is a leading global technology company and strategic life cycle partner to the pharmaceutical, biotech and food industries. Headquartered in Stuttgart, Germany, it employs 7,300 people at 49 locations in more than 20 countries. With over 160 years of industry expertise, Syntegon delivers solutions that help people live better, stay healthier and trust the products they rely on every day.

The Challenge: Scaling AI responsibly #

Syntegon needed to scale AI access across a global, multi-site engineering organization while keeping strict data governance, compliance and full control over AI models and data sources. Its leadership set a goal: give every employee enterprise-grade AI capabilities without sacrificing security and compliance.

Syntegon found that public SaaS AI tools posed unacceptable risks around data residency and IP exposure. The company also found that commercial SaaS tools offered:

• too little control over data residency,
• no way to integrate proprietary internal data sources securely,
• limited choice of models, and
• no enterprise-grade role-based access aligned to Syntegon’s identity management.

Given how strategically important AI is to the company’s future, vendor lock-in was also a major concern.

Syntegon’s non-negotiable requirements were:

• Deployment in Syntegon-controlled cloud infrastructure (AWS, EU region)
• SSO with Microsoft Entra ID, plus role-based access through IAM
• Support for multiple LLM providers (Azure OpenAI and AWS Bedrock) through a unified gateway (LiteLLM)
• A Work Council and data protection officer (DPO) approval pathway that complies with local regulation
• A structured AI literacy training program that every employee must complete, covering both the strategic benefits of AI and the responsible boundaries of its use

The Solution: Syntia on AWS #

Open WebUI’s open architecture, permissive licensing and enterprise-ready features let Syntegon rapidly deploy a fully self-hosted, enterprise-hardened AI platform without customizing the codebase. Those features include RAG, Model Context Protocol (MCP), SSO, multi-model support and an extensible plugin framework.

LiteLLM proved decisive. Integrated as a unified, cost-controlled gateway, it routes traffic across Azure OpenAI, AWS Bedrock and future AI models. Open WebUI’s configurability also lets Syntegon’s security teams enforce strict governance policies, including disabling code execution and blocking unauthorized webhooks.

• Compute and orchestration: Amazon EKS (managed Kubernetes) on AWS Fargate
• Storage and database: Amazon Aurora (PostgreSQL-compatible) with the pgvector extension for RAG embeddings; AWS EFS for persistent shared storage across pods; daily AWS RDS snapshots
• Networking: All application workloads run in private subnets of an AWS VPC, with no direct inbound internet access to pods
• Secrets: AWS Secrets Manager for all sensitive credentials, and AWS KMS for encryption at rest
• CI/CD: GitOps-based deployment, with Terraform for infrastructure as code and Kubernetes manifests managed in Git. Planned updates are tested in staging before production, with documented rollback plans.
• Logging and monitoring: An OpenTelemetry Collector for traces and metrics, Prometheus for metrics storage, Grafana dashboards for visualization and alerting, AWS SNS for email alerts, and the LiteLLM dashboard for per-model usage and cost
• Security controls: MFA via the identity provider, device compliance, role-based access by group, enforced data residency, moderation and guardrails, audit exports to SIEM, and egress restrictions

Models and data handling #

All model routing runs through a LiteLLM gateway deployed inside Syntegon’s Kubernetes cluster.

• Azure OpenAI: GPT-5 for high-quality tasks; GPT-5-mini as the default for cost-effective, high-volume tasks; GPT-4.1 for complex reasoning; GPT-5.4 and GPT-5.1; and GPT 1.5 for image generation
• AWS Bedrock: Claude Sonnet 4.5 and 4.6 for advanced reasoning and long context; Pixtral Large 25.02 for vision and text; and the open-source gpt-oss-120b

GPT-5-mini is every user’s default, and users can choose other models from the full catalogue. Cost-sensitive or high-volume tasks go to GPT-5-mini or gpt-oss-120b. High-sensitivity or complex reasoning tasks can use Claude Sonnet or GPT-5.

RAG uses Azure OpenAI text-embedding-3-large with pgvector on Aurora, which Syntegon prefers for reliability across multiple pods. Sources include:

• internal knowledge bases uploaded to the Syntia Workspace,
• SharePoint document libraries, through the Microsoft Graph API,
• Confluence pages, through the Atlassian integration.

Syntia handles PDFs, Office files, CSV, Markdown, HTML, JSON, images and audio. OCR and Azure Document Intelligence extract content from complex documents.

Sensitive data is protected in four ways:

• Azure AI Content Safety filters all Azure model inputs and outputs, and AWS Bedrock Guardrails apply equivalent filtering on AWS models.
• Admin chat access and admin exports are disabled to protect user privacy.
• Direct signup is disabled. Users sign in only through Entra ID SSO and are provisioned through the Syntegon Identity Manager.
• Daily backups are encrypted at rest.

Adoption and Enablement #

Syntia went live in production in December 2025, with access provisioned through an Azure Service Group. It has been updated iteratively since. Employees in eligible regions can onboard themselves through the Syntegon Identity Manager.

Syntegon supports adoption through:

• Training and documentation: dedicated Syntia training materials and e-learning modules, with Confluence and SharePoint as the single source of truth for Syntia documentation, including architecture, configuration, security and rollout guides.
• AI Hub & Spoke sessions: a central hub team owns the shared AI infrastructure, governance, approved tools and reusable components, while departments act as spokes that apply that foundation to their own problems.
• Product Board meetings: used to gather feedback, prioritize features and share updates. An AI Innovation Hub coordinates these meetings, use-case ideation and roadmap governance across departments.
• Support and community: office hours, functional team enablement sessions and a dedicated AI Hub community.

Syntia is used across engineering and R&D, operations and manufacturing, project management, sales, marketing and customer experience, finance and HR, and digital transformation. It supports German, English and other languages through Azure Speech multilingual neural models. LiteLLM virtual API keys let engineering teams build their own AI-powered applications on Syntia’s model infrastructure.

Results #

“We moved from AI experiments to real productivity gains, with more than 1,000 users in the first weeks of go-live — not quarters. Open WebUI let us operationalize use cases at scale while keeping full visibility into cost and compliance.”

— Matti Maier, Director of Digital Technology Applications, Syntegon

Syntegon reports the following results:

• Security and compliance: Syntia is the only AI platform at Syntegon approved under the company’s data governance and Work Council framework. It has two layers of content filtering (Azure AI Content Safety and AWS Bedrock Guardrails), a full audit trail through OpenTelemetry and Grafana, and daily snapshots that meet its backup and recovery SLAs.
• Enterprise knowledge: More than 500 active RAG agents give context-aware answers grounded in Syntegon product documentation, SharePoint and Confluence.
• Ecosystem integration: Syntia is the only enterprise AI platform at Syntegon with native integrations to SharePoint, Confluence, Jira, web search, image generation and document RAG in one interface. That removes the need for multiple siloed tools.
• Model breadth and cost efficiency: 8+ models across Azure and AWS Bedrock in a single interface, with routine tasks going to GPT-5-mini and premium models reserved for complex work.
• Platform reliability: Syntia runs on a production-grade, autoscaling Kubernetes cluster with a 99.9%+ availability target. Planned maintenance has full rollback plans, and no unplanned downtime incidents have been documented.

“The biggest win with Open WebUI is having every LLM in one place. Our teams switch between models for reasoning, coding, or RAG-based retrieval on the fly—one interface, full flexibility, no juggling separate tools or accounts.”

— Matti Maier, Director of Digital Technology Applications, Syntegon

SharePoint Search and Confluence Search appear in Syntia’s tools menu.

Top Use Cases #

Syntegon highlights these use cases and benefits:

• Enterprise knowledge at your fingertips: 500+ RAG agents answer questions from product documentation, SharePoint and Confluence, with no manual searching
• Developer productivity: integrated code assistance and review speed up delivery cycles and improve code quality
• Customer and field service: AI-assisted brief drafting for faster, more consistent service communications
• Visual content generation: on-demand image generation and editing for technical illustrations and marketing assets, reducing external dependency and turnaround time
• Agile at scale: the Confluence integration supports sprint planning, documentation and project tracking

Why Syntegon Chose Open WebUI #

• Open and self-hosted: an open architecture and permissive licensing, deployed fully self-hosted without customizing the codebase
• Enterprise-ready features: RAG, MCP, SSO, multi-model support and an extensible plugin framework
• Unified model gateway: the LiteLLM integration, routing across Azure OpenAI, AWS Bedrock and future models, which Syntegon called decisive
• Configurable governance: security teams can enforce strict policies, such as disabling code execution and blocking unauthorized webhooks

What’s Next #

Working with its partner, Syntegon could build an advanced integration that pulls data from Confluence and SharePoint while honoring each user’s access permissions, so users retrieve only the content they’re entitled to see.

Looking for an Enterprise Plan? — Speak with Our Sales Team Today!

Get enhanced capabilities, including custom theming and branding, Service Level Agreement (SLA) support, Long-Term Support (LTS) versions, and more!