Harvey Expands Legal AI With Memory, Review Automation and Cross-System Search

Harvey rolled out a September product update that combines user-specific memory, agent-led document review and broader enterprise search across its legal AI platform. The September 16 release targets three persistent adoption barriers in legal technology: inconsistent outputs, labor-intensive document analysis and fragmented access to firm knowledge. It also extends Harvey’s reach into Microsoft Word, iManage, identity systems and multiple research sources.

The update shows Harvey moving beyond standalone question-and-answer use cases toward a more persistent operating layer for legal work. Its product suite already spans research, document storage, contract analysis, collaboration and organizational analytics. The latest changes connect those functions more tightly, allowing preferences, permissions, templates and source material to travel with a user or matter. That direction reflects a broader shift in enterprise legal AI from isolated tools toward workflow systems that can participate in repeatable processes.

Users can now save drafting or negotiating preferences and apply them across Harvey interactions. A new template capability can generate documents, client presentations and memoranda in a firm’s preferred format. Custom agents can also run inside Word, while the Word add-in lets lawyers compare alternate clause language and revisit earlier recommendations from conversation history. For large-scale review, Harvey added a redesigned table workflow: users can describe the fields they need, ask the system to reorganize or edit rows, and locate existing tables through a dedicated search interface. The platform can also search an entire Vault or Space, identify relevant files and cite the pages supporting its response.

Research capabilities expanded in parallel. Harvey said users can combine as many as five knowledge sources in one query and added more than 150 legal research sources, including material from Chilean publisher Tirant lo Blanch. An iManage integration brings natural-language precedent search to Harvey’s web application and Word add-in, with citations returning to source documents. Client-matter controls now include Playbook assignments, searches by name or description, and required acknowledgment of a client’s AI policy. Synchronization through Entra ID or Okta can keep job titles, practice areas and locations current for Command Center reporting.

Industry Implications

The release places workflow continuity, rather than model novelty alone, at the center of competition among legal AI vendors. Enterprise legal buyers will increasingly assess whether systems preserve permissions, provenance and firm-specific preferences across applications. For legal operations teams, the review-table functions could matter as much as the underlying model because they target the structured work that often determines whether AI delivers measurable time savings. Integrations with Word, iManage and identity providers also raise the bar for rivals: adoption depends not only on answer quality, but on fitting existing technology and governance environments.

DreamLegal Perspective

Legal technology vendors should measure success by how reliably their systems carry context through a matter without weakening source traceability or access controls. Legal operations professionals evaluating AI agents should test memory, citations, permissions and human review together, not as separate features. Harvey’s September release suggests that the next competitive phase will be decided by workflow depth and administrative fit, with product breadth becoming valuable only when users can control how the system acts across their existing stack.