Lidia is an artificial intelligence platform for legal professionals that supports the full cycle of legal work, from research and analysis to drafting and verification. The platform is built on Italian legislation, case law, and a firm's own document base, using specialist AI agents to execute multi-step legal tasks and citing outputs back to their source material.
Key features and functions include:
Specialist AI Agents
Dedicated agents handle each stage of legal work: research, analysis, drafting, and verification, carrying out multi-step actions such as researching primary sources, reviewing contract clauses against a firm's playbook, drafting contracts and pleadings, and verifying citations and statutory references.
Legal Research
Agents identify and analyze official statutory and case law sources using GraphRag technology, selecting sources based on relevance, hierarchy, and coherence with the legal query.
Web Research
Research can be extended to selected external online sources when needed, with traceability maintained back to the source.
Smart Answer
A natural-language query function for the firm's document repository that returns answers with references to the specific documents and passages used, which can then be carried into further analysis or drafting.
Workflow Builder
Allows firms to construct and reuse customized AI workflows that replicate their own professional processes and methods.
Microsoft Word Add-in
Lidia can be queried directly within Word documents to analyze or integrate content, or to request edits shown in markup/track-changes mode.
Case File Organization
Documents can be grouped into case files by client, matter, or deal, giving a consolidated view across research, analysis, and drafting for a given matter.
Document Context and OCR
Each document retains up to 1,000 pages of context for analysis, and an OCR function converts PDFs, scans, and other complex files into text the AI can process.
Data Security and Privacy
Client data is segregated by tenant, encrypted at rest and in transit, and not used to train or improve underlying LLM models.
Loading...