Introducing Harvey II: smarter agents from the start. - Featuring Harvey Tenet, our first model trained for legal work - Built around matters and projects - Agents start with the files, context, permissions, and history they need - Assign tasks to lawyers or agents, then track and review the... work - Harvey remembers how you work and writes like youshow more

Explore the new Harvey:

Should have called the new model Louis

the part that gets harder as an agent inherits more context is enforcing the ethical wall, not the drafting. firms silo matters for conflict reasons, and an agent pulling history across projects needs that same barrier logic, or it becomes the leak the wall exists to prevent.

just to be clear, youre going to post train off client data and sell the use the model back to them? locked-in so they cant run it themselves?

I love the design style behind this!

Is Harvey named after the tv show Suits?

The “trained for legal work” sounds weighty, innovative, and helpful. Trained with what? Does this now mean that every Harvey Tenet AI output will be watermarked under the EU law?

Didn’t you start out by finetuning your own models and gave up when general models got 🥲 ? Is this more investor bait or something changed from back then?

A strong step toward truly context-aware legal AI agents

How does the matter management work for in-house teams?

“Harvey has worked through the latest batch of material contracts in the data room overnight.” Agentic work is becoming more asynchronous and latency-tolerant; human verification and accountability remains the bottleneck.

Legal AI that actually knows your matters Harvey II is a game changer

Agents that start with the files and context they need that's how you actually save time!

Harvey II is pushing legal AI from chatbots toward true workflow agents.

Harvey Tenet + matter-based agents is such a smart leap — agents that actually start with full context is huge!

to the moon

seen legal agents leak context when matter permissions go stale, access must refresh per task

Harvey II agents starting with legal context and memory is huge

Building agents around legal context makes them far more practical

💪🏻

The interesting part is not the model, it is the layer above it: matters, permissions, review. That is where a firm's method gets encoded, and it is why vertical beats generic assistants.

whose legal work ? your own customer's ?

Harvey 2.0 🥰

We’re still in the early stages of the AI revolution. The biggest opportunities may not even exist yet.

looks really useful for real legal work

@harvey When an agent starts with files, permissions, and history, how do you show a lawyer exactly which context drove its answer? That trace may matter as much as the answer.

First model trained just for legal