Harvey

What is Harvey?

Harvey is a specialized legal AI platform rather than a general-purpose chatbot. It helps professionals research legal and regulatory material, analyze uploaded and connected documents, draft and revise work product, extract information from large document collections, and run repeatable workflows.

A user can begin with a prompt, a document, a connected knowledge source, a Vault, or a configured workflow agent. Harvey then produces a cited answer, draft, summary, review table, report, transcript, translation, or other work product that can be checked and revised by a professional.

Its target users include law firms, corporate legal departments, legal operations teams, litigation groups, transactional lawyers, and tax or finance professionals handling confidential information. Unlike general assistants such as ChatGPT or Claude, Harvey is organized around governed legal work, document collections, institutional knowledge, and repeatable professional workflows.

How legal teams use Harvey

Research, drafting, and document analysis

Harvey’s Assistant supports questions over uploaded files, legal knowledge sources, and connected systems. It can summarize material, answer questions with supporting citations, draft or revise legal documents, translate content, and work from precedents, templates, or internal instructions.

This makes it useful for tasks such as preparing a first draft, comparing provisions, reviewing a contract, analyzing a regulatory issue, or turning source material into a structured memo. The resulting work still requires legal judgment and review; Harvey does not remove the need for professional verification.

Vault and large-scale review

Vault is Harvey’s workspace for organizing and analyzing large document collections. Teams can ask targeted questions across a document set, generate bulk summaries, and create review tables that extract defined fields from many documents for comparison.

This is particularly relevant to due diligence, transaction review, litigation preparation, regulatory investigations, and other matters where reading documents one at a time is inefficient. Harvey documentation states that a Vault can store up to 100,000 documents, although practical limits and available features can depend on the workspace configuration, file type, permissions, and agreement.

Workflow Agents and Agent Builder

Workflow Agents support multi-step processes such as contract review, diligence, regulatory analysis, drafting, translation, and litigation preparation. Agent Builder allows an organization to configure agents using firm knowledge, playbooks, connected systems, and defined review or approval steps.

The distinction from ordinary chat is important: the goal is not only to generate an answer to one prompt, but to make a recurring process more consistent and easier to supervise. Teams can use agents for defined workflows while keeping people involved in review and approval.

Integrations and working environments

Harvey is designed to sit alongside existing legal and business systems. Reported integrations and connected sources include Microsoft Word, Microsoft Outlook, SharePoint, OneDrive, Microsoft 365 Copilot, Google Drive, Gmail, Box, iManage, NetDocuments, LexisNexis, EUR-Lex, EDGAR, Aderant iTimekeep, and other connectors. Availability can vary by workspace, geography, permissions, and rollout status.

Microsoft Word and Outlook access matters because it allows legal work to remain close to the documents and communications where it already takes place. Harvey also supports Spaces and governed collaboration for sharing work with internal teams, clients, or other authorized collaborators.

Native iOS and Android applications extend access beyond the desktop. Depending on the enabled features, users can query Vault, scan documents, dictate prompts, and transcribe calls. These mobile capabilities complement the core document and research workflows rather than turning Harvey into a general mobile assistant.

Models, knowledge, and governance

Harvey publicly documents services and model providers including OpenAI, Anthropic, Google, Mistral, Microsoft, DeepL, ElevenLabs, Baseten, and Fireworks.ai. It also documents selectable GPT-5-series, Claude, Gemini, Mistral Medium 3.5, and Harvey Fable models. Exact availability depends on the workspace, region, feature, rollout stage, and organizational configuration.

The platform’s value for enterprise teams is therefore not just model access. It combines model selection with knowledge sources, permissions, auditability, workflow configuration, and connections to legal information and document-management systems. Administrators can manage access through role permissions and SAML single sign-on, with audit logs, IP allow-listing, and other enterprise controls documented by the company.

Privacy and data considerations

Harvey states that customer data and content are not used to train its AI models or improve its products and services. It also states that its AI-provider subprocessors are contractually prohibited from training on customer data or content. Customer data is described as encrypted in transit and at rest, logically separated between customers, and protected by access controls.

Retention is governed by the customer agreement and workspace settings. Harvey states that model providers generally use zero data retention, but some optional or covered AI services may have different storage, retention, or review terms when enabled. Regional processing options may be available, and customer administrators control access, sharing, retention settings, and optional features.

These policies are relevant for legal organizations, but they do not eliminate the need for due diligence. Buyers should review the applicable agreement, data-processing terms, subprocessors, regional requirements, and settings for each enabled integration or AI service.

Pricing and access

Harvey does not publish standard self-service subscription pricing or a public starting price. Access is generally arranged through an enterprise or organizational agreement and a request-a-demo process. A custom evaluation period may be available, but no standard public trial duration was verified.

Commercial terms may vary according to users, features, integrations, usage, deployment requirements, and workspace configuration. This structure makes Harvey more suitable for organizations evaluating a governed legal platform than for individuals looking for an inexpensive, immediately self-serve chatbot.

Strengths and limitations

Where Harvey fits well

• Law firms and corporate legal teams working with confidential, document-intensive matters.
• Due diligence, contract review, regulatory analysis, litigation preparation, and legal research.
• Organizations that need review tables and bulk analysis across large document sets.
• Teams seeking repeatable agents connected to playbooks, knowledge bases, and approval processes.
• Enterprises that require access controls, SSO, auditability, regional processing options, and integrations with existing systems.

Important limitations

• Public pricing and standard plan details are unavailable, so cost cannot be assessed without an organizational sales conversation.
• Setup, administration, integrations, and governance can require substantial enterprise work.
• Feature, model, connector, and regional availability may differ between workspaces.
• Legal outputs can contain errors or omissions and require qualified professional review.
• Some optional or covered AI services may have different data-processing and retention terms.
• Harvey is specialized for professional-services workflows and is not intended to be a low-cost general-purpose assistant.

Is Harvey a good fit?

Harvey is a strong candidate for organizations that need legal AI embedded in controlled research, drafting, document-review, and workflow processes. Its combination of Vault, review tables, knowledge sources, workflow agents, and enterprise integrations is more relevant to a legal department than a standalone chat interface.

It is less suitable for casual users, individuals who want transparent monthly pricing, or teams that only need occasional general writing assistance. Buyers should also be prepared to assess security terms, configure permissions, connect approved systems, and establish review procedures before relying on generated work in consequential legal matters.