Lagora automates legal due diligence with multi-model AI agents
Stockholm-based Lagora has deployed an AI agent platform that processes up to one million cells of legal data simultaneously, aiming to shift law firm billing from hourly rates to project fees. The company’s multi-model strategy routes tasks to specific LLMs based on cost and capability, with recent updates improving citation accuracy from 70% to 99%.
Lagora, a legal tech company founded in Stockholm and backed by YC 24, has developed an AI agent platform designed to automate high-volume legal tasks such as due diligence and contract review. The platform features a 'Tabular View' that allows users to process up to one million cells in a single table, extracting data from thousands of agreements simultaneously. To satisfy law firm security requirements, Lagora hosts its infrastructure on Azure, leveraging Microsoft’s status as an approved vendor for legal data. The company employs a multi-model strategy, routing tasks to different large language models based on cost, latency, and capability. For example, Gemini is used for transcription and speaker attribution, while post-trained open-weights models handle specific retrieval tasks. Lagora recently acquired Cura and Wexler to expand its capabilities in M&A and litigation fact intelligence. CTO Jacob Lauritsen stated that the company is hiring in New York, Stockholm, and London, with a current team of 100 engineers. Lauritsen argues that AI will not eliminate junior lawyers but will polarize the profession, making skilled lawyers more valuable while automating menial tasks. He believes law firms will shift from hourly billing to project fees for lower-end work, focusing human capital on strategic judgment and client relationships. A dedicated evaluation team of three engineers and three lawyers tracks model performance across legal practice areas. They monitor metrics such as citation accuracy, drafting quality, and red flag identification. A recent model release improved citation accuracy from 70% to 99% without changes to Lagora’s harness. The company uses a model selector in its agent interface, allowing users to choose between models like Fable, Opus, and Astra, though most legal work does not require the highest-tier intelligence. Lauritsen noted that while 60% of legal work can be handled effectively by current models, complex analysis still requires higher intelligence.