Personalize Harvey with User-Level Memory

Overview

User-level Memory is now available, allowing you to save personal preferences, such as drafting style, negotiating posture, jurisdiction, or formatting habits, simply by telling Harvey to remember them as you work. Once saved, your preferences apply automatically to every new session.

Key Features

• Save preferences via Assistant conversation: Tell Harvey to remember a preference in any Assistant thread, no separate setup step or form needed.
• Automatic preference detection: Harvey can recognize when something you've said reflects a preference or fact about you and save it automatically, so you don't have to say "remember this" for it to stick.
• Persistent memory across sessions and surfaces: Saved memories apply automatically in Assistant on the web app, add-ins (Word and Outlook) and Playbook runs.
• Manage everything in one place: View, edit, and delete every saved memory from your Settings page.
• Global on/off toggle: Turn Memory on or off anytime in Settings.

How To Enable Memory (Admins)

1. In Settings, select Roles.
2. Enable the User Memory permission for relevant roles.

How To Use Memory (Users)

Save a Memory

1. In any Assistant thread, tell Harvey what to remember. For example: "Remember that I represent the buyer," or, "Always apply Delaware law." You don't always have to ask explicitly: Harvey can also recognize a a preference you'd likely want applied again and saves it automatically.
2. Review the in-thread nudge Harvey surfaces, which shows exactly what was saved.
3. Select edit or delete on the nudge to adjust or remove it on the spot.

Apply Memories

Start a new thread and run a relevant query. Your saved memories apply automatically across the web app and add-ins (Outlook, Word).

When a memory is applied to a response, Harvey shows a memory citation so you can see which memory was used. These citations are attached to either specific parts of the answer or the entire response. They are visible only to you: if you share a thread with others, they can see the answer but not the memory citations behind it.

Manage Memories

1. In Settings, select Memory to open your memory table.
2. From the table, you can:

   1. view all memories
   2. edit any memory's text
   3. delete individual memories
   4. create a memory directly without going through a thread

Turn Memory On or Off

In Settings, use the Use Memory in responses toggle to turn Memory off. When toggled off, nothing is saved or recalled.

Use Cases

Memory can be used to save a wide range of preferences, including:

• Interaction preferences: How Harvey should respond. For example, “Be less verbose,” or “Use bullet points.”
• Common legal positions or defaults: For example, “We usually accept X, push for Y.”
• Drafting preferences: How work product should be produced and formatted. For example, “Always use active voice” or “Avoid complex formatting.”
• Facts about you: Where you practice, where you're barred, your practice area, client types, things you look out for on tasks.

Notes and Limitations

• Memories are capped at 1,000 characters of text.
• A user may have up to 1,000 memories saved at once.
• Review Tables and Workflow agents do not currently save or recall memories.
• Duplicating a shared thread does not carry over the original user's memories, apart from those already captured in the thread.

FAQs

Q: Can I ask Harvey to remember specific information about my work?

Yes. You can ask Harvey to remember specific details about how you work, such as your preferences, formats, and recurring instructions. Memories are stored at the user level and are not associated with a space, client, or matter. You are responsible for what you save and when to use it.

Q: Should I save client names or confidential details in memory?

No. Because memories are stored at the user level and are not yet separated by client or matter, keep client names and confidential details out of memory. Save only light, non-confidential preferences. Matter-level memory, which stores memories by client or matter, is coming soon and will make it easy to associate a memory with a specific matter.

Q: How does retention interact with memory? What happens when the source thread is deleted or expires?

When you save a memory, it exists independently of the thread it came from and does not have a retention period, just like a saved prompt in your library. Deleting or expiring the source thread does not delete the memory.

Q: How do security, data, and privacy work for Memory?
Memories are text snippets stored within your existing Harvey environment, under the same security posture as the rest of Harvey, and are not used to train models.

Each user's memories are private to that user, not shared with other users or pooled across your organization. Memories persist until you delete them and are not subject to your workspace's standard retention policy. You can view the full privacy policy on User Memory in the Harvey Trust Center.

Q: Can admins see users' memories?
No. Memories are private to the user. Admins can enable or disable the permission but cannot view or edit anyone's memories.

Q: What happens if my prompt conflicts with a saved memory?

The prompt wins. While saved memories serve as helpful background, Harvey treats your current prompt as the most relevant. If your request directly contradicts something a memory, you might occasionally notice Harvey reasoning through the difference before landing on your instruction.

If you’d like a memory or category of memories to be ignored for a query, you can simply say so in your prompt.

Q: When should I use a writing style vs. memory?

Use a writing style when there are formats or tones to apply consistently in Harvey across a team or applied firm-wide. Think of it as "publishing" a standard once so everyone can use it. See our article on custom writing styles for more details on creating & sharing

Use memory for preferences that are personal to you, that you'd rather not re-select every time, or that emerge naturally out of how you work day to day - things you'd otherwise have to remember to specify in every prompt.

Q: Does user-level memory replace writing styles?

No, we see the features as complementary. Writing styles remain the best option for formal, shareable, firm-approved formatting standards. User-level memory is best for the ongoing, personal preferences that can be saved to streamline future threads.

Q: What happens when a memory conflicts with my writing style?

Harvey will generally prioritize the language in your prompt over memories or writing styles in the event of conflicts.