OneAdvanced deploys 50+ AI agents on UK-sovereign AWS infrastructure

OneAdvanced faced a fundamental tension: its customers in healthcare, legal, and other regulated sectors demanded strict UK data residency and privacy controls, yet the AI models and managed services the company wanted to deploy were not available in UK regions at the time. Rather than compromise on data sovereignty to access managed AI services, OneAdvanced chose to self-host open-weight models on AWS infrastructure it controls entirely, running on p5.48xlarge GPU instances in the London region. This decision shaped every layer of the architecture—from model serving through vLLM, to agent orchestration on Amazon ECS, to a custom RAG pipeline built in-house. The company's move from prototyping with Amazon Bedrock (which showed rapid results in a two-week sprint) to self-hosted deployment reflects a broader shift: organizations handling sensitive data can now build production-grade AI systems without relying on centralized cloud AI platforms, as long as they have the engineering capacity and AWS account access to do so.

The speed of agent deployment—over 50 agents in three weeks, most in under a day—suggests that the technical and operational barriers to rapid agentic development have fallen significantly. OneAdvanced's choice of Strands Agents SDK over competitors like LangChain and LangGraph points to a model-first, low-ceremony approach to agent definition: each agent is defined with a system prompt, a tool set, and an optional input form, then containerized and deployed. The no-code agent builder extends this accessibility to non-developers, allowing product managers, clinicians, and business analysts to author agents without writing code. This democratization of agent creation, combined with a shared tool library (calculator, web search, Snowflake integration, UK statute law search, and others), enabled OneAdvanced to move from concept to production rapidly.