How Claira and Harvey Work Together

Your firm may already own Harvey. If it does, the question in front of you is not which tool wins. It is which part of the matter each one carries, and what has to pass between them for the handoff to be worth anything.

That question comes up on almost every call we take with a litigation team that has already bought into legal AI. There is a Harvey licence, a partner who has built real fluency with it, and a discovery population that has just crossed six figures. Someone asks whether Harvey can simply handle the review, and the useful answer is more specific than a yes or a no.

Harvey and Claira sit at different points in the same matter. This post is about where the seam falls and how to work it.

What Harvey Does Well

Harvey ships four products: Assistant for chat, drafting and document analysis; Vault for storing and analysing document sets in bulk; Knowledge for legal and regulatory research against cited sources; and Workflow Agents for building repeatable multi-step tasks without writing code. Vault syncs folders directly out of iManage, SharePoint and Box, which means the documents a matter team has already curated are one connection away from being queryable.

For the work lawyers do around discovery, that is a strong hand. Harvey is good at reading a pleading and telling you what the plaintiff actually has to prove. It is good at turning a hundred contracts into a review table with the indemnity and termination provisions extracted. It is good at drafting the memo, at answering a regulatory question with citations you can check, and at building a workflow that runs the same analysis over every new document that lands in a folder.

We are not going to argue with any of that. If your team has invested in Harvey, keep investing. The output is genuinely useful, and much of it is useful precisely because it sits upstream of the job Claira does.

Where the Handoff Has to Happen

The boundary is not intelligence. It is location and volume.

Vault carries a per-vault document ceiling, and that ceiling is sized for curated matter materials rather than for a raw collection. This is the correct design decision for the work Harvey is built around. A deal room, a regulatory production you have already narrowed, an investigation file with a few thousand documents in it: those fit the shape, and Harvey handles them well.

A discovery population does not fit that shape. After culling, a twenty-custodian collection can still leave several hundred thousand documents that each need a defensible decision recorded against them. Those documents live in Nuix Discover, with family relationships intact, the coding panels your reviewers already work in, a privilege log maintained alongside the review, and a production workflow everything eventually has to pass through.

Lifting that population out of Nuix Discover and into a separate workspace is not a change of workflow. It is a migration, and migrations of evidence come with custody and completeness questions you will be asked to answer later. The more sensible move is to leave the population where it is and bring the review to it. Claira runs inside Nuix Discover, reads each document a reviewer would have read, applies your criterion, and writes both the decision and the reasoning back into the coding fields your QC process already trusts.

Turning Harvey's Work Into Case Context

This is where Harvey's output stops being a document and starts being operational.

Case Context is a structured summary of the matter that lives inside Claira and shapes the prompts it generates. It is organised into five sections: parties and people, description and timeline, relevance and issues, privilege indicators, and collection details. Every one of those maps onto something Harvey can produce for you in an afternoon.

Ask Assistant to pull the party and entity list out of the pleadings, including subsidiaries, trading names and every individual named in the statement of claim. That becomes parties and people, and it doubles as the vocabulary Claira watches for. Ask for a chronology of the events in issue, with each date anchored to the document that supports it. That becomes description and timeline. Ask it to reduce the pleadings to a numbered issues list. That becomes relevance and issues, and it is the section that does the most work once a bulk scan is running.

Two cautions are worth stating plainly. Case Context is visible to everyone on the case, so keep work product and mental impressions out of it, however tempting it is to paste in the strategy memo Harvey drafted. And keep it short. It is a briefing, not a case memo, and length tends to dilute it rather than sharpen it. We have written before about how to translate a case theory from Harvey or CoCounsel into a working Claira prompt, and the same discipline governs here.

Sending Coded Evidence Back Up

The handoff runs in both directions, and the return leg is the one teams tend to forget.

Once a bulk scan has finished, you have a coded population rather than an undifferentiated one. The documents flagged as hot, or as going to a particular issue, are a set measured in hundreds rather than hundreds of thousands. You know which they are, you know why each one was coded that way, and you can sample and defend the decision.

That set is exactly the size Harvey is good at. Export it, load it into a Vault, and put Assistant to work on the narrative: the deposition outline, the chronology of what the defendant knew and when, the argument in the summary judgment brief. The evidence has already been found and coded defensibly inside Nuix Discover. Harvey then does what it does best, which is help you build something out of it.

Running Both Without Paying Twice

The failure mode we see is rarely tools competing. It is two teams reading the same documents twice because nobody wrote down where one job ends and the next begins.

Set the rule at the opening of the matter. Anything requiring a decision recorded against every document in the population is Claira's job and it happens inside Nuix Discover. Anything requiring synthesis across a small, deliberately chosen set is Harvey's job. Research, drafting and client-facing work product stay with Harvey from beginning to end.

Then hold the vocabulary steady. Issue names, code names, deal terms and acronyms should read identically in the Harvey issues list, the Claira coding fields and the eventual privilege log. That consistency is not cosmetic. It is what lets you walk opposing counsel through your process without improvising, and what lets you justify the AI line on the bill when the client asks about it.

Neither tool is trying to become the other. Harvey is building the lawyer's workspace. We are building the reviewer's, inside the platform the review already runs in. The teams that get the most out of both are the ones that decided early which floor each job belongs on.

If you want to see what that handoff looks like against your own collection, book a short demo with our team. We will take a criterion from one of your live matters, run it across a sample, and show you the coded output, the reasoning behind it, and exactly where your existing Harvey artifacts would slot in.