Harvey is an enterprise legal AI platform where cited agents, a bulk document Vault and legal research sources support law firms and corporate legal departments; access is sold through demos and order forms, not public plans.
Used this tool? Rate it
Used this tool? Rate it
Harvey is a legal AI platform for law firms and corporate legal departments, which the company positions for contract analysis, due diligence, compliance, and litigation. It combines task-running agents, a bulk document repository called Vault, a legal research layer called Knowledge, and shared workspaces called Spaces. Access is sold to organizations through demos and order forms rather than a public sign-up page.
Adoption figures come from the vendor and were captured on September 28, 2026. Harvey says more than 200,000 lawyers around the world use the platform. Its September 2026 funding announcement adds that 80% of Am Law 100 law firms use Harvey, alongside in-house legal teams at five of the Fortune 10. These are company statements rather than audited counts.
The company, frequently searched as Harvey AI, was founded in 2022. Chief executive Winston Weinberg is a former lawyer who co-founded it with Gabe Pereyra, previously a research scientist at Google DeepMind and Meta. Business press coverage gives a sense of scale: Harvey reached $190 million in annual recurring revenue in January 2026. In September 2026 it raised $550 million at a $15.5 billion valuation. The same coverage reports that Harvey's first in-house model, Harvey Tenet, was built from the open-weight Kimi K3 model and post-trained with legal data.
Harvey's modules share one document layer: vault content is available across Assistant, Workflows, Mobile, and the Word and Outlook add-ins, so the same files can be queried from several surfaces.
Memory works alongside the agents. Harvey says it remembers how a user works and applies those preferences automatically across sessions and surfaces, which targets the same problem as Agent Builder from the individual side: fewer repeated instructions for recurring work.
Knowledge is the research layer. Harvey describes it as connecting each query to legal databases, curated public sources, and a firm's own institutional knowledge. Named database partners include Ask LexisNexis, Gyldendal Rettsdata, FromCounsel, SCC Online, Wolters Kluwer, Otto Schmidt, Lefebvre, and Tirant lo Blanch. The company also cites a data pipeline into 500+ legal sources across jurisdictions. Its source list also covers EU law, regulations, and Court of Justice decisions. EDGAR covers SEC filings and public company disclosures. For AI legal research, that means answers can draw on licensed databases and public regulatory material instead of only a general web index.
A typical agent-assisted matter in Harvey moves from documents to a checked deliverable:
The approval step and the citation log are the two controls Harvey offers for keeping a lawyer responsible for the result.
Harvey's own examples are concrete legal tasks rather than general productivity claims.
It is not a consumer legal-help service. Harvey's websites and services are not directed to anyone under 18, and there is no self-serve sign-up path on the pages checked; individuals looking for advice on a personal matter are outside its stated audience.
One mobile detail differs between sources. The Harvey Mobile product page presents transcription of multi-party calls or hearings as available now. The App Store listing, by contrast, marks Audio-to-Transcript as "Coming Soon". The two pages were captured on September 28, 2026 and do not agree.
Harvey pricing is not published. No price list or self-serve plan appears on the product pages or sitemap as captured on September 28, 2026, and the contact page asks prospects to book a demo. Commercial terms are set per customer:
| Access route | What the terms say | Scope |
|---|---|---|
| Paid subscription | Fees and payment terms are set in each customer's order form | Features covered by the order form |
| Evaluation (free trial) | Use runs through an authorized free trial, and Harvey may limit trial resources and features | "Basic Service" for general legal research |
| Add-on functionality | Covered separately in an order form or other writing | Custom workflows, third-party integrations, certain research modules |
The evaluation terms define the Basic Service as general legal research in which users provide input and receive output. Harvey can end trial access immediately if it believes a trial is not being used in good faith.
Refunds under the platform agreement are narrow. If an order form is terminated for an uncured material breach before its end date, the customer receives a pro rata refund of prepaid, unused fees. The agreement also lets Harvey correct invoicing errors within 45 days of an invoice being received.
Independent legal-technology coverage names two platforms as Harvey's closest competitors:
The practical difference for a buyer is where the legal content comes from. CoCounsel is built on Thomson Reuters' own research products, while Harvey relies on outside data partners to unlock curated and proprietary knowledge. Because Harvey publishes no list price, a cost comparison with either rival depends on quotes.
These limits put responsibility for accuracy on the lawyer. Citations are meant to make checking easier, but the terms themselves say output can be wrong or incomplete.
No model training by default: the platform agreement states that Harvey will not train any AI models using customer content or customer data, and neither will its subprocessors.
Model providers: Harvey contractually prohibits model providers from training on customer data and requires Zero Data Retention from them.
Opt-in custom models: a firm can have its client data used for training only on explicit request, in which case a bespoke model is created exclusively for that firm.
Retention controls: customers decide what data to upload, how long it is retained, and whether it can be shared with others in the company.
Deletion after termination: within 30 days of termination, Harvey deletes remaining customer data or content unless instructed otherwise.
Hosting and regions: Harvey hosts its cloud environment in Microsoft Azure and offers processing in the EU and Switzerland or Australia for customers with data localization needs.
Privacy policy scope: the website privacy policy does not apply to inputs, outputs, or uploaded documents; Harvey processes those as a data processor, and questions go to the customer as data controller.
AI regulation: Harvey's AI policy commits it, to the extent applicable to its service, to comply with AI laws including the EU Artificial Intelligence Act.
These are vendor-described commitments and audits. Whether they satisfy a specific firm's confidentiality or regulatory obligations is outside what the published pages establish.
No public free plan exists. Evaluation happens through an authorized free trial under separate terms, limited to the Basic Service for general legal research, and paid access is priced in individual order forms.
Not by default. The platform agreement says Harvey and its subprocessors will not train AI models on customer content or data. A firm can request a bespoke model trained on its own data, used only by that firm.
The terms say output is not legal advice and may contain errors, and the acceptable use policy requires review by a licensed professional before any licensed service, such as legal advice, is provided.
Harvey runs on Microsoft Azure and offers data localization with processing in the EU and Switzerland or Australia, in addition to the US.
Yes. It has Word and Outlook add-ins, an email channel, native iOS and Android apps, DMS integrations, and APIs/MCP for custom integrations.
