Memory is Here: :Harvey:, Personalized

Every lawyer has their non-negotiables. The Oxford comma is essential, or it's clutter. "Shall" is sacred, or it's archaic and struck on sight. Anglo-Saxon rooted words preferred to Romance-rooted words. Until now, AI made you re-teach these preferences in every prompt.

Memory, released as part of Harvey II, changes that. It closes the gap between how you practice and how Harvey executes — automatically, transparently, and always under your control.

Standing Instructions, Not Repeated Prompts

A memory is like a standing instruction that applies itself: Once set, it applies to all your work moving forward. Getting started takes just a few questions about how you work. After that, you can tell Harvey to remember something in the middle of what you're already doing. For example:

• The GC I email reads on his phone. Keep email drafts to him under 150 words, with the ask in the subject line.
• When there is a judgment call, give me two options — a conservative 'safe harbor' path and an 'aggressive' path — each with a risk level.
• Always distinguish the UK GDPR position from the EU GDPR position. Never blend them into a generic “GDPR” answer.
• I am admitted only in New York. Flag anything that requires Delaware or other local counsel input.

Save memories one at a time or capture several at once. The more you tell Harvey, the more it adapts — and the more your work comes back the way you'd do it yourself.

Saved Once, Applied Everywhere

A preference saved once travels with you across the Web App, Outlook, and the Word Add-In, in playbook runs, and every agent you put to work. Tell Harvey how you like your partner or GC updates written. Harvey maintains those standards everywhere you draft updates, whether in a doc or an email, with no reminder and no reformatting.

This is what makes agents more useful, not just faster. An agent that carries your context does every step of the work your way: your framing, your positions, your format. Run a dozen agents at once, and each one will work your way. That means more work completed, all to your standard, giving you time back for higher-value work.

And wherever a memory is applied, you can see it. When a memory shapes a response, Harvey cites it in it the same way it cites a source. You can see which preferences were used, confirm it’s the right call, and catch any instances that need correction.

You Control What Harvey Remembers

General-purpose AI assistants treat memory as a background process: the model quietly accumulates what it learns about you, and you find out what it stored, or how it's being used, only if you go digging. That's fine for consumer chat. It's below the bar for legal work.

Harvey is built on the opposite principle: you always know what it remembers, when it learned it, and when it's being used.

Harvey will save memories when you explicitly ask (“Remember to…”) and when a prompt signals a preference (“I want all summaries to start with BLUF”). In both cases, a confirmation card shows exactly what was saved so you can control what Harvey stores. Every memory lives in one place in a user’s Personal Settings, where they can review, edit, search, or delete any memories, or turn Memory off entirely.

Admins control whether Memory is available in their workspace, including whether to enable it for all users or a select subset. And it's protected by the same security protections as everything else in Harvey. Memory opens no new access to your systems, and like the rest of a customer’s workspace content in Harvey, individual memories are never used to train AI models.

Make Harvey Yours

“:Harvey:'s Memory feature lets me spend less time re-establishing context and more time doing sophisticated work. It remembers my role, drafting style, analytical approach, and output preferences, so I can focus on better prompts, deeper follow-up questions, and higher-quality legal analysis.”

“An organization’s institutional context lives with its people: their expertise, insights and understanding of what matters most to their business and customers. We have worked with :Harvey: to shape Memory, which carries this growing context forward so users can build on it each time they use the tool. Critically, this context is stored securely, appropriately governed and being used in the right way. ”

“Working with :Harvey: on Memory early on, we emphasized that users need full transparency and control over what is stored and used — a non-negotiable requirement for legal work. It’s great to see that principle reflected in how :Harvey: has designed Memory. We're looking forward to seeing our lawyers spend less time repeating context and get closer to a final draft from the first prompt.”

“We actively partnered with :Harvey: on Memory because it enables our people to capture their preferences, standards and drafting approach once, then apply them consistently across matters and engagements. One of the most powerful unlocks is in its ability to understand how we each work, not just respond to what we ask. Personalisation like this makes AI feel like a genuine extension of our lawyers, rather than just another tool. ”

Since announcing Memory in January, we've shaped it with the teams who will rely on it — AmLaw and Magic Circle firms, Fortune 500 and FTSE legal departments — across the US, UK, Europe, the Middle East, and Asia-Pacific.

Memory today is scoped to you alone: your preferences, in your workspace, visible to no one else without your knowledge. It's our introduction to memory as a concept in Harvey, and it's the foundation for what comes next. We're already building memory scoped to a shared space, so context will stay tied to the right client matter or project, with organization-level memory to follow.

Memory is available in Early Access today, with general availability in the next couple of weeks. Reach out to your Harvey account team or contact us for a demo.