npx @buildinternet/releases get harveyRecently shipped Harvey Graph, which grounds agent reasoning in a structured map of a user's documents and data.
Harvey Graph grounds agents in your own corpus. Agents now reason across a structured graph of the documents and data a firm already has, rather than working from isolated files.
Motions to dismiss now come out fully cited. Harvey drafts a researched, citation-backed motion to dismiss from a complaint, grounded in the LexisNexis collection of U.S. case law, statutes, and Shepard's Citations Service.
Connectors are vetted up front and governed at runtime. The Connector Library curates pre-verified connectors for agents; an MCP policy engine governs tool access, information flow, and agent actions beyond that initial review.
Search and sync reach more systems.
Legal research widened by 130+ sources. Octus adds leveraged finance and restructuring content, and Knowledge Source filters are now savable and defaultable, including web search. Excel files are read with cell formulas, saved values, and cell-level citations.
Still shipping from prior weeks. Custom Harvey Agents run inside Microsoft Word, and contract review agents carry lessons from past deals into new ones.
Horizon Scanning lets users monitor regulatory and legislative developments and act on them directly in Harvey without tracking sources by hand.
Harvey for Outlook now supports additional knowledge sources, bringing firm documents, jurisdictional research, and public web results into email queries.
Harvey's Connector Library is now available, bringing team tools into Harvey via security-reviewed connectors or custom MCP.
Harvey Graph improves agent accuracy by grounding agent reasoning in a structured graph of the user's documents and data.
Workspace audit logs can now be viewed, filtered, searched, and exported directly in Command Center without an API integration.
Harvey expanded legal research coverage with more than 20 new data sources.
Harvey adds a new Agents page and redesigned sidebar for quicker access to agents, recent work, and Spaces, plus an updated Library for managing reusable resources from a single place.
Workspaces with client matters enabled can now tag Spaces to an individual client matter.
Custom Agents let teams turn their processes and standards into reusable agents that can take on routine work.
Harvey can now create email drafts directly in the Outlook Drafts folder from threads in the web app, the Harvey for Word add-in, and agents, in addition to the Harvey for Outlook add-in.
Custom agents and workflows can be used as soon as they are saved, with admin approval applied only at the sharing levels a firm chooses.
Harvey now features a refreshed visual design built on a single, consistent design system, applied across Harvey II, Review Tables, and the broader user experience.
An in-app Notifications hub shows everything that needs attention across Harvey and links directly to the relevant task, comment, or resource.
Vault files can now be reorganized with plain-language instructions from a new thread in Harvey.
Harvey can now suggest the right sources for a request, including vaults, review tables, knowledge sources, connectors, and custom agents, and can deliver multiple outputs from a single request.
A single New button now starts threads and review tables from a Vault, and new threads stay private in your History until you add them to the Vault.
Several new models are now available in Harvey: Opus 5.5, Sonnet 5.5, GPT-6.1 Sol, GPT-6 Sol, and GPT-6 Luna.
Harvey now automatically suggests a custom agent when your request matches an agent available to you.
Spaces has been rebuilt as the central home for all of a lawyer's work on a matter or project, spanning collaboration and sharing.
Harvey's monthly roundup introduces a curated Connector Library with admin controls, natural-language Agentic Search across SharePoint and Google Drive, and Gmail search from within a Harvey thread. Also included: PacerPro Vault Sync, Ironclad file imports, a LexisNexis motion-to-dismiss workflow, 130+ new legal research sources, Fable 5.1 and GPT-6 Astra models, mobile agents, and localization in German, Italian, and Portuguese.