Understanding Retention Policies

Learn how data retention works in Harvey, including how the Workspace Retention Period applies and which features follow their own retention timelines.

Last updated: Jul 22, 2026


Overview

At Harvey, we prioritize giving customers control over their own data. Each customer can set a Workspace Retention Period — the amount of time data is retained in Harvey after it’s created or uploaded.

Your workspace retention can be customized anywhere from 3 hours after the data is processed by Harvey to 30 days after your agreement with Harvey ends.

As Harvey’s product evolves, some features require their own retention timelines to maintain peak performance, reliability, and functionality. Furthermore, as our LLM partners continue to enhance their LLM offerings, different retention timelines may apply.

This article explains which features have separate retention rules and why.


Workspace Retention Period

Your Workspace Retention Period determines how long data in your workspace, such as Assistant queries, is stored before it’s automatically deleted. It is measured from each query’s creation date and does not reset when a query is updated.

Once your workspace reaches its configured limit, content is securely and permanently removed from Harvey’s systems, unless the feature has its own specific retention configuration described below.


Features with Independent Retention Timelines

Some features are designed to function independently of the workspace retention setting to prevent data loss or preserve essential configurations.

Learn more about Vault specific retention in Manage Vault Data Retention Settings.

Feature

Retention

Why it's Different

Vaults

  • Admins can configure Vault Retention, separate from Workspace Retention.
  • Retention can be set from 3 hours to indefinite, measured from the project’s creation date or last activity date; projects auto-delete at the end of the period unless deleted sooner.
  • Flexible Vault retention settings allow you to retain important work beyond the workspace period, preventing accidental loss of important work.

Custom Workflows (including Words to Workflow)

  • If not deleted, they are automatically deleted 30 days after the Agreement ends.
  • Custom workflows are retained independently to ensure teams don’t lose their automations once workspace retention ends.
  • Note: Inputs and outputs produced by workflows still follow workspace retention.

Library

  • Admins and users can delete starred inputs and examples at any time.
  • If not deleted, they are automatically deleted 30 days after termination.
  • Libraries are stored beyond workspace retention to avoid automatic deletion of valuable references or examples.

Playbook Reference Guide and Rules

  • Admins and users can delete these at any time.
  • If not deleted, they are automatically deleted 30 days after termination.
  • Retaining playbooks and rules ensures guidance and compliance materials remain accessible beyond workspace retention.

Matter Numbers

  • Admins and users can delete matter numbers at any time.
  • If not deleted, they are automatically deleted 30 days after termination.
  • Matter numbers support consistent reporting and historical tracking for your records, even after the workspace retention period ends.

User Profiles

  • Admins and users can delete profiles at any time.
  • If not deleted, they are automatically deleted 30 days after termination.
  • Keeping user profiles temporarily allows for personalized experiences during offboarding and transition periods.

Custom Export Formats

  • Admins and users can delete export formats at any time.
  • If not deleted, they are automatically deleted 30 days after termination.
  • Retaining export templates prevents the need for admins to reconfigure templates every time the retention period expires.

Extended AI Features

  • Admins can turn on Extended AI Features. The retention will depend on the specific feature.

  • [For Claude Fable 5] Customer Data and Content may be retained by the model provider for at least 30 days. See Extended AI Features for more.

  • [For Claude Fable 5] If any data or content is flagged by a classifier for safety investigation or Anthropic is legally required to retain it, such data or content may be retained beyond 30 days.

Prompt Caching

  • The key/value (KV) tensors — the numerical “memory” the model produces during the first tokens of a prompt — may be stored for up to 24 hours, depending on the model provider.
  • Key/value tensors are stored so subsequent, similar prompts can skip that costly pre-fill step.

Prompt Caching

Harvey continues to enhance its offerings as its LLM partners develop and launch new AI models.

To stay up to speed on the models Harvey evaluates and deploys for customers, see What AI Models Does Harvey Use, which covers Harvey's model evaluation process and a comparison of current models.

For more on how Harvey approaches prompt caching as these models evolve, see our Prompt Caching overview.


FAQs