Important Note: The Vault feature is not included in the base Harvey application. If you’re interested in adding Vault to your plan, please reach out to your Harvey Success Manager for more details.
Overview
Vault is a powerful tool that enables you to upload hundreds of documents to a project and conduct advanced analysis using Harvey’s AI models. It’s especially valuable for workflows that involve reviewing large document sets.
To see a list of projects, or create a new one, click into the Vault section in the left sidebar of Harvey.
Create a Project
Before you create a project, keep these helpful guidelines in mind:
Each user can create up to 10 Vault projects within their workspace.
Each project can store up to 10,000 files with a total capacity of 5GB, providing plenty of space for your needs.
Individual files can be as large as 100MB, while Excel files have a 20MB size limit to ensure proper data extraction.
Vault supports .doc, .docx, .pdf, .xlsx, .xls, .eml, .csv, .msg, .txt and .pptx file types.
Click the 'New Project' button in the top right corner of the Vault page.
Enter a Project Name.
Select documents you wish to upload.
Click 'Create Project' in the bottom right corner.
Use a Project
First, select the desired project from your Vault tab. Here, you can run a new query, view recent queries, upload new files, delete existing files, and organize files into folders.
When creating a new query in Vault, you have two options:
Create from blank:
Review Query: Get individual answers from each document in a table format.
Assistant Thread: Get a single, aggregated answer across all documents.
Workflows: Instantly execute predefined queries tailored to specific use cases, such as M&A Deal Points or Litigation Court Opinion workflows.
Assistant Threads (Ask Query)
You can perform Ask queries on Vault projects from either Vault or Assistant, with the same result: an Assistant thread generated from a Vault project.
In Assistant, select 'Choose from Vault'. For detailed instructions on running a query with Vault files, please visit our Assistant article.
In Vault, select 'Create blank' and then 'Create thread'.
Queries from either method will show up in the Vault project as well as in your History, indicating that Vault is the source.
Review Table
With the updated review grid creation, you can create and manage columns individually. Each column has a type, which is the kind of data expected in the cell. Review queries extract an answer that matches the column type.
Column Types: date, duration, currency, number, verbatim, classify, and free response.
Custom Sorting: columns can be sorted and formatted based on the data type.
Export: you can export the whole table or sub-selected rows into an Excel sheet.
Document View: if you click a document, you can view each answer and relevant sources.
Workflows
Execute common use cases without typing a single letter. Right now, Harvey offers seven workflows across transactional and litigation practice areas:
Sharing Projects
Users can share projects in Vault with other users within your workspace. Sharing is restricted to members of the same organization, ensuring that only colleagues in your workspace can access shared items.
For more details on how to share projects, please visit our Sharing Vault Projects article.
Delete or Rename a Project
From the Vault tab:
Hover over the project you wish to change.
Click on the ellipsis menu in the top right corner of the project.
You will have the option to either 'Rename project' or 'Delete' the project.
If you choose to delete your project, you will be prompted again to confirm this decision as this action will permanently delete the project and all associated sub-folders and files. Once a project is deleted, it cannot be recovered.
Data Security & Privacy
Important note: Documents uploaded in Vault are categorized as ‘Customer Data’ in your platform agreement.
Data Storage Buckets
Vault projects and data adhere to the specific workspace and territory-level data storage buckets assigned by Admins. For example, if data is saved in an EU storage bucket, Vault files will also be stored there.
Data Retention Periods
There are two types of data retention periods related to Vault: query retention and projects & data retention.
Query Retention: The default retention period for Vault queries is 48 hours, unless your firm's retention period is longer. This ensures that users can access and export their Review queries, which can take 20-40 minutes to run, before they are deleted.
Projects & Data Retention: Vault projects and data remain stored until the user deletes them, independent of workspace or territory-level retention policies. However, data will be deleted if your organization ceases to be a Harvey customer.
Important Note: Once Vault projects, queries, files, or data are deleted, they cannot be recovered by Harvey or the customer.
FAQ
Q: Why don't I see Vault in my navigation pane?
This is likely because Vault is not part of your workspace configuration or plan. Please contact your Harvey Admin to explore how you can gain access.
Q: Does a user's uploaded documents in Vault help "train" Harvey's AI for other customers?
No, all data uploaded to Harvey only helps improve outputs for the user who uploaded it. Harvey does not retain data to train its AI models.
Q: How many files can users select from a Vault project?
You can select the entire project — up to 10,000 documents. Querying Vault documents in Assistant is not subject to the 50 document limit that Assistant has for uploaded documents.
Q: How does Vault differ from using Assistant with a large number of documents?
Vault enhances Harvey’s ability to store and efficiently process large document sets, offering several advantages:
Targeted Querying: Easily query specific sub-folders and files within a large document set.
Smart Document Categorization: Automatically route documents by type (e.g., NDAs, leases) for more accurate and relevant responses.
Optimized Search for Specific Information: Significantly improves performance in solving ‘needle-in-the-haystack’ problems, such as locating a specific name across thousands of documents.
Structured Responses: Provides individual answers from each document in a clear, table format for easier analysis and comparison.
Q: What are the review evaluation benchmark's for Vault?
We’ve evaluated Vault on a variety of use cases and datasets across transactional diligence, litigation, deal term/precedent tracking, IP and other practice areas.
Datasets evaluated include:
Patents
Emails
Transactional agreements
Case summaries
Commercial contracts
Miscellaneous documents — Government documents, regulatory filings, etc.
Vault performs between 94%-99% accuracy on objective Review tasks with 10-50 questions. (An “objective task” refers to a query that can be verified as true or not.)
For further assistance, please contact [email protected]