OneAdvanced Deploys 50+ AI Agents on UK-Sovereign AWS

OneAdvanced builds UK-sovereign AI with self-hosted models

OneAdvanced, a UK enterprise software provider serving over 10,000 customers in healthcare, legal, and other regulated sectors, has deployed more than 50 specialized AI agents on AWS infrastructure that keeps all data within UK borders. The company self-hosts Llama 4 Maverick and Llama Guard 4 models on Amazon SageMaker AI to meet strict data sovereignty requirements that many managed AI services couldn't satisfy at the time of implementation.

The solution went from prototype to production serving customers in July 2025 and has been running for over a year, according to details first reported by AWS.

Why it matters

Data sovereignty is no longer a nice-to-have for enterprises in regulated industries—it's a hard requirement. OneAdvanced's approach demonstrates that organizations can deploy cutting-edge AI capabilities without compromising on data residency, a critical consideration for public sector and healthcare customers who cannot accept data leaving national boundaries. The company achieved ISO 42001 certification for AI governance, positioning itself among the first UK organizations to meet this standard.

Architecture and implementation

OneAdvanced runs Llama 4 Maverick in FP8 precision and Llama Guard 4 on p5.48xlarge GPU instances in AWS's London region. The company moved from p4d.24xlarge instances to P5 hardware specifically to support 120,000 to 128,000 token context windows needed for large document analysis and extended conversations. They use vLLM for model serving and leverage reserved instance discounts to manage GPU compute costs.

The agent layer runs on Amazon ECS, with each of the 50+ agents configured with its own system prompt, tool set, and optional structured input forms. Agent configurations live in Amazon DynamoDB. OneAdvanced built these agents using Strands Agents SDK after evaluating LangChain, LangGraph, and other frameworks. The company went from its first agent to 50+ deployments in three weeks, with most individual agents built in under a day.

For document retrieval, OneAdvanced built a custom RAG pipeline they call "Llamadex" using Amazon Aurora PostgreSQL with the pgvector extension. Documents uploaded to Amazon S3 are converted to markdown, chunked into 2,048-token segments, and embedded using the intfloat/multilingual-e5-large-instruct model. The system supports both personal and organization-level document collections with full isolation.

No-code agent builder expands access

OneAdvanced created a visual agent builder that lets non-technical users—product managers, clinicians, business analysts—create and configure agents without writing code. Users define agent personas, design input forms with drag-and-drop fields, write system prompts that reference form values, and select from a shared tool library that includes calculator functions, chart creation, Snowflake database queries, UK statute law search, and web search capabilities.

Some agents use an interview-style pattern, asking structured questions to build context through turn-taking rather than expecting users to provide complete information upfront. OneAdvanced's Strategic Thinking assistant, for example, guides users through three specific questions before delivering advice.

Security and guardrails

Llama Guard 4 screens every user input for harmful content before requests reach the main inference model. OneAdvanced replaced an earlier Llama Guard 3 deployment after observing high false rejection rates. No user queries or responses are retained or used for model training, and uploaded documents remain inaccessible even to OneAdvanced engineers. Amazon GuardDuty provides threat detection on the S3 storage layer.

Andrew Henderson, CTO of OneAdvanced, emphasized in the company's AI launch video that data sovereignty is "a hard requirement for many of our customers, especially those in the public sector and highly regulated industries."

The AWS advisory engagement that supported this deployment received a 5 out of 5 customer satisfaction score. Alex Savage, Head of Integration at OneAdvanced, noted the company "learned a lot which helped us launch a unique offering to the market."

These implementation details were first reported by AWS in a blog post co-authored with the OneAdvanced team.