Delivering agentic AI to law firms globally using AWS with stp.one

Founded in 1993, stp.one is a leading provider of AI-based legal tech software in Europe. Even while faced with talent shortages and evolving regulatory requirements, law firms are managing more and more cases, each involving critical deadlines for filing motions and responding to requests. With the ongoing advance of agentic AI, the company saw an opportunity to address these challenges.

“Most of the work that legal professionals do is 99 percent document-driven,” says Dirk Eisenberg, chief technology officer at stp.one. “Some cases have timelines of more than 10 years, and that’s where Legal Twin can help. It can process tons of documents to build a concrete timeline of a case, which lawyers and paralegals can use to better prepare for court.”

Serving notaries and law firms, stp.one’s software must comply with the GDPR as well as Germany’s federal code for lawyers and other data residency requirements. “We needed to take care that the whole chain of regulations are fulfilled from us, as the vendor, down to the customer,” says Eisenberg. “We know that AWS is a trustworthy hyperscaler and knew that we could build a framework that would meet our market regulations.”

To develop AI-powered capabilities, stp.one adopted Amazon Bedrock, which provides its teams with access to hundreds of foundation models from leading AI companies. By using Amazon Bedrock, stp.one’s developers have the greatest model choice and flexibility, empowering them to quickly build transformative experiences for their clients. “Amazon Bedrock was the perfect fit for us,” says Eisenberg. “We wanted the ability to select different models and control what data we use with the data privacy features that Amazon Bedrock offers.”

The company developed six unique AI-powered capabilities for Legal Twin, which range from intelligent contract risk analysis to automated receivables processing. To generate documents and answers using natural language processing, stp.one uses Anthropic’s Claude in Amazon Bedrock.

The company built Legal Twin on top of the document management system STP Documents, which uses Amazon Aurora PostgreSQL to power semantic searches with high performance and availability at global scale for PostgreSQL. To perform semantic searches across its cloud-based vector store, stp.one selected the Titan embedding model in Amazon Bedrock.

By using STP Documents as its knowledge base, Legal Twin’s responses rely on case documents or court decisions, which results in higher accuracy and increased reliability. To further build trust with its clients, Legal Twin includes citations for its generated content. “The user can see what court decision is being referenced,” says Manuel Naujoks, tech lead at stp.one. “That way lawyers can themselves verify the information that’s being provided.”

Legal Twin also features flexible integration options. Clients can embed the solution directly into their workflows using the Model Context Protocol (MCP), an open-source standard for connecting AI applications with external systems.

Clients that must meet stringent data retention and data sovereignty requirements can also choose to store and process all their data locally. “We are the only company that makes it possible for notaries to use cloud-based AI capabilities,” says Eisenberg. “We can connect up on-premises storage solutions and cloud-based processing, which means that notaries can take advantage of AI features without having to purchase additional GPUs.”

Legal Twin is helping law firms around the world accelerate their legal research, litigation discovery, case records analysis, document retrieval, invoicing and collections, and more. For example, certain claims traditionally take 6–10 minutes to file, and by using Legal Twin, users can process those same claims in under 30 seconds.

Further, stp.one’s fully managed strategy has freed up its developers to focus on releasing software updates and new features. “We’re currently working in 2-week sprint cycles, which means that our customers get something useful from us at least every other week,” says Eisenberg. “Sometimes it’s a small change, and sometimes it’s a bigger user experience change. We’re always looking at ways to improve our clients’ workflows.”

The company will continue to enhance Legal Twin’s ability to integrate with other AI agents and complex workflows. “We’re shifting from AI assistance to a multi-agent system where agents can interact with each other and delegate work,” says Naujoks. “With powerful support from AWS, we’re able to safely experiment with these agents to keep pace with the changing times.”