Harvey
Harvey is domain-specific AI for legal and professional services, used by 2,400+ law firms and in-house teams across 70+ countries for contract analysis, due diligence, compliance and litigation work. Enterprise only, sold through a demo request.
What is Harvey?
Harvey is a legal AI platform built by Harvey AI Corporation and described by its makers as domain-specific AI for legal and professional services. It is aimed at law firms and corporate legal departments, and its privacy policy widens the audience to professionals in law, tax and finance. The company reports 2,400+ organisations, more than 200,000 professionals, 75+ AmLaw 100 firms and use in over 70 countries.
The platform is organised into modules. Agents plan and execute complex legal work end to end, routing tasks automatically, running several agents in parallel, working on a schedule, accepting documents, images, video or audio, and returning reports, presentations or interactive pages. Agent Builder lets a firm encode its own playbooks and preferences, and memory carries context from one task to the next. Vault stores, organises and bulk-analyses large document sets. Knowledge answers legal, regulatory and tax questions, drawing on a claimed 500+ regional knowledge sources. Spaces opens shared workspaces across firms, clients and co-counsel while enforcing the firm's existing ethical walls, which Harvey applies but never creates, modifies or deletes. Contract Intelligence speeds routine review, supports negotiation and surfaces portfolio-wide insight, and Command Center reports adoption by team, practice group and office with external benchmarking. A newer generation, Harvey II, is advertised on the home page.
Everything is designed to be checkable: every claim carries a citation and every step is logged. Harvey publishes usage figures of 850,000+ queries and 50 million+ files processed daily, 700,000+ agent tasks a day and 25,000+ custom agents in production. Named customers include Deutsche Telekom, Bayer, A&O Shearman, Repsol, Ashurst, Cuatrecasas, Reed Smith, Syngenta and PwC UK.
It reaches users through regional web workspaces, native iOS and Android apps, Word and Outlook add-ins, an email channel, and an API with MCP support. There is no self-serve route: the product is sold through a demo request and no price is published anywhere on the site.
What it does
- Delegate a complex legal task to an agent that plans it, runs it end to end and returns a cited, review-ready deliverable
- Store, organise and bulk-analyse thousands of contracts, filings and client documents in Vault
- Research legal, regulatory and tax questions across domains and get answers grounded in trusted sources
- Draft and redline documents inside Microsoft Word using natural-language instructions and firm playbooks
- Summarise long email threads and their attachments, then generate client-ready replies inside Outlook
- Share a governed workspace with clients and co-counsel while existing ethical walls stay enforced
- Track adoption, usage and benchmarks by team, practice group and office in Command Center
When to use Harvey / When not to
A quick filter to help you decide if Harvey is the right fit.
When to use Harvey
- Large law firms running client matters across litigation, transactional and regulatory practices on one governed platform
- Corporate in-house legal departments managing contract portfolios and outside counsel relationships
- Mid-sized firms with lean teams that need agents to absorb high-volume review and drafting work
- Legal operations and innovation leads who must drive adoption and measure impact across offices
- Compliance, regulatory and tax specialists who need cited, source-grounded answers on fast-moving rules
When not to use Harvey
- Solo practitioners and small firms: there is no self-serve signup, no published price and no free tier
- Anyone wanting to try before committing, since the only entry point is a sales demo and a signed order form
- Teams handling payment card workflows, as the service is explicitly not PCI compliant
- Use cases involving genetic or biometric identifiers, which the Platform Agreement forbids as inputs
- Users located in countries under United States embargo, and anyone under 18
How to use Harvey
A typical end-to-end flow, from setup to results.
- Request a demo on the contact-sales page, since there is no self-serve signup
- Scope the deployment with the sales team and sign an order form under the Platform Agreement, an Evaluation Terms of Service covering pilot phases
- Choose the processing region, EU and Switzerland, the United States or Australia, a choice that also binds Harvey's subprocessors
- Configure enterprise access: SAML single sign-on, IP allow-listing, audit logs and data retention policies
- Sync the firm's existing ethical walls so Harvey enforces them while your walls provider stays the system of record
- Connect the document management system, whether iManage, NetDocuments, SharePoint or Google Drive, along with tools such as Aderant or Ironclad
- Sign in through the regional workspace at app.harvey.ai, eu.app.harvey.ai or au.app.harvey.ai
- Work day to day by querying the Assistant, uploading documents to Vault, running a workflow or delegating to an agent
- For agent work, preview the plan, adjust the scope and approve before it runs, then review the cited output before it leaves the firm
- For custom integrations, create a bearer token in the workspace API Tokens settings and call api.harvey.ai/v2 or its EU and AU endpoints
Pros & Cons
Pros
- Contractual data guarantees that can actually be checked: a No Training clause passed down to subprocessors, Zero Data Retention required of model providers and an enforceable Security Addendum
- Real data residency choice between the EU and Switzerland, the United States and Australia, applied to subprocessors and mirrored by regional app and API endpoints
- Every claim carries a citation and every step is logged, so work can be verified before partner or client review
- A published, dated subprocessor list with an objection right and a notification sign-up
- A deep certification stack covering SOC 2 Type II, ISO 27001, ISO 27701, ISO 42001, CCPA and the Data Privacy Framework, with named external auditors
- It meets lawyers where they already work, inside Word, Outlook, email, mobile, the document management system and an API
- Adoption at a scale few legal tools reach: 2,400+ organisations, 200,000+ professionals and 75+ AmLaw 100 firms
Cons
- No public pricing at all: a complete 667-URL sitemap contains no pricing page, so budgeting is impossible before a sales conversation
- No free plan and no advertised free trial, the only way in being a demo request
- Firmly enterprise-shaped, with a master agreement, an order form and a Customer Success Manager standing between the buyer and the product
- The mobile apps are described but not distributed publicly, with no App Store or Google Play link anywhere on the site
- No interface language list is published, which is awkward for a product sold in more than 70 countries
- Outputs are contractually framed as a research tool rather than legal advice, so human review stays mandatory
- Terms can be updated unilaterally, and the 15-day notice period does not cover changes Harvey judges non-material
Pricing & Plans
Harvey publishes no pricing. Neither a free plan nor a free trial is advertised, and no rate card exists on the site: a complete review of the 667-URL sitemap returned no pricing page, and no amount appears anywhere in the collected material. Fees are set individually in each customer's order form under the Platform Agreement, so no lowest price point can be stated in any currency. Access begins with a demo request, while an Evaluation Terms of Service governs evaluation and pilot phases. Billing questions go to accountsreceivable@harvey.ai, invoicing errors may be corrected within 45 days of receipt, and fees are quoted exclusive of taxes and government assessments.
Data, GDPR & hosting
A consolidated view of how Harvey handles your data.
GDPR overview
Implementation is concrete and documented. Harvey displays a GDPR badge on its home, platform, agents and security pages, and publishes a privacy policy effective 13 July 2026 alongside a Data Processing Addendum, a Data Transfer Addendum and a dated subprocessor list carrying an objection procedure. A data protection officer is designated under Article 37: Dr. Felix Wittern of Fieldfisher in Hamburg, reachable at DPO@harvey.ai, with privacy@harvey.ai as the general contact. An EU entity, Harvey AI Ireland Limited, is established in Dublin. The policy sets out its legal bases, lists data subject rights and details transfer mechanisms: adequacy decisions under Article 45, the June 2021 Standard Contractual Clauses under Article 46(2), and Article 49 derogations. Harvey is certified under the EU-U.S. Data Privacy Framework, its UK Extension and the Swiss-U.S. DPF, and holds ISO 27701. No Article 27 representative is named, the Irish establishment making one unnecessary.
Who owns the data?
The customer owns its data. Under the Platform Agreement the client keeps all right, title and interest, intellectual property included, in Customer Data, and the text is explicit that as between the parties you own your Content. Harvey receives only a non-exclusive, worldwide, royalty-free licence to process that material as far as needed to deliver the service, address technical problems or comply with applicable law. For uploaded documents, queries and responses Harvey acts as data processor and the customer is the controller, so privacy requests about that material are directed to the customer. Harvey is controller only for account, communication and browsing data described in its privacy policy.
Reuse rights
Harvey commits contractually not to reuse customer material for model development. The No Training clause states that Harvey will not train any AI model on the client's Content or Customer Data, and the same prohibition is passed to subprocessors which, cloud storage providers aside, may not retain or log that material for human review. Model providers are held to Zero Data Retention. One exception exists and it is opt-in: a firm may explicitly request a bespoke model trained on its own material, which is then never used for other customers. Separately, Harvey collects Usage Data, the metadata generated by interaction with the service, to develop, improve, support and operate it; usage data carrying the client's confidential information may only be shared under the confidentiality clause or in aggregated, anonymised form. On the customer's side, outputs are the client's to reuse without asking permission. The agreement does grant Harvey the right to name the client as a reference and use its logo, a right the client may terminate at any time on notice.
Data retention & training
Hosting summary
Harvey hosts its cloud environment on Microsoft Azure. Customers with data localisation requirements can choose processing in the EU and Switzerland, in the United States, or in Australia, and that choice extends to Harvey's subprocessors. Regional application workspaces and matching API endpoints exist for each: app.harvey.ai, eu.app.harvey.ai and au.app.harvey.ai. The published subprocessor list gives the locations of each provider, with Microsoft, OpenAI, Google Cloud Platform and Amazon Web Services spanning all four zones, Mistral AI confined to the EU, and Anthropic, LexisNexis and the web browsing providers in the United States. Data is encrypted at rest and in transit, logically separated per customer, and access follows least privilege. Transfers outside the country of origin rely on adequacy decisions, the 2021 Standard Contractual Clauses or Article 49 derogations, and Harvey is certified under the EU-U.S. Data Privacy Framework, its UK Extension and the Swiss-U.S. DPF.
Where Harvey works
Country-level availability.
Not available in
Things to keep in mind
Risks and trade-offs to weigh before adopting Harvey.
- Outputs are AI-generated and contractually described as a research tool that may contain errors, misstatements or gaps, so signing off without reading them is the core professional risk
- Delegating drafting and research at scale can erode junior lawyers' craft, since the tasks agents absorb are precisely the ones trainees used to learn on
- Training on customer data is off by default but can be switched on by explicit request, a decision that deserves internal governance rather than a single signature
- The service is not PCI compliant, genetic and biometric identifiers are forbidden as inputs, and protected health information requires a separate Business Associate Addendum
- Harvey enforces ethical walls but never creates or owns them, so an out-of-date walls provider will see its wrong boundaries faithfully applied
- Terms may be updated unilaterally with 15 days' notice, and less for changes Harvey judges non-material, leaving periodic re-reading to the customer
- The agreement grants Harvey the right to name the customer as a reference and use its logo until that right is actively revoked
Setup & Integrations
Technical difficulty
Two very different levels. Deployment is an enterprise project: a commercial process, a signed order form, then SAML SSO, IP allow-listing, audit logs and retention policies to configure, a processing region to choose, ethical walls to sync from the firm's existing provider, and connectors to wire into the document management system and business tools. Daily use asks almost nothing technical, since it happens inside Word, Outlook, email and a mobile app. Support is substantial, with embedded legal engineering teams, Harvey Academy and a help centre. API work needs a token created in workspace settings, sometimes only by an administrator.
Deployment
Integrations
Behind Harvey
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
How much does Harvey cost?
Is there a free plan or a free trial?
Does Harvey train its models on my data?
Can I have a model trained on my firm's own data?
Where is my data hosted?
Who owns the documents I upload?
Does Harvey have an API?
Are there mobile apps?
Which security and privacy certifications does Harvey hold?
Is there a minimum age?
Should you pick Harvey?
Harvey is one of the few legal AI products that has clearly outgrown the pilot stage. The adoption figures it publishes, 2,400+ organisations and 75+ AmLaw 100 firms across more than 70 countries, put it in a category of its own, and the platform has broadened from a research assistant into an agent system that plans work, executes it in parallel and returns cited, review-ready output.
Its strongest argument is not the feature list but the paperwork behind it. The No Training clause is contractual rather than promotional, it binds subprocessors as well as Harvey, model providers are held to Zero Data Retention, and the subprocessor list is published and dated with a genuine objection right. Data residency in the EU and Switzerland, the United States or Australia is real and extends to subprocessors. Add SOC 2 Type II, ISO 27001, ISO 27701 and ISO 42001, named external auditors and an enforceable Security Addendum, and the security story holds up to the scrutiny a general counsel will give it.
The weakness is commercial opacity. Nothing about price is public, there is no free tier and no way to try the product without a sales conversation and a signed order form. That is a defensible choice for enterprise software, but it makes Harvey impossible to evaluate independently and rules it out for solo practitioners and small firms.
The honest summary: if your organisation can contract and run a deployment, Harvey is a serious candidate with unusually verifiable data commitments. If you are looking to test legal AI on your own, this is not the door. And whatever the tier, the agreement is blunt that outputs are a research tool and not legal advice, so the professional judgement stays with the lawyer.
- Choosing a selection results in a full page refresh.
- Opens in a new window.