Harvey raised 500M for legal AI, I built the same and open sourced it | Hacker News Reader

Harvey raised 500M for legal AI, I built the same and open sourced it

github.com

69 pointszztopisgreat38 comments

> CaseCite is a _source-available_, self-hosted legal RAG caselaw research & document analysis platform.

Source-available is not open source.

Especially with clauses like this:

> You may not move, change, disable, or circumvent the license key functionality in the software, and you may not remove or obscure any functionality in the software that is protected by the license key.

Doesn't matter. The entire thing is vibe coded, so you can just point Claude at it and say "Copy this".

You can do that even if it wasn't vibe coded. What a time to be alive!

Well, technically yes, ethically no.

I feel its probably unreasonable but I do get a sense of stolen valour when I click on "I built..." and can immediately tell its vibe coded.

Claude will now refuse to do that ... which makes my blood boil due to how Claude itself was built.

Recent commit...

LICENSE | Relicense under the Apache License 2.0 | 7 minutes ago

too slow, already flagged

A lot of Harvey’s value prop is not in the underlying software - someone else came out with an “open source Harvey” (called “Mike”) months ago and it hasn’t really put a dent in uptake of Harvey or its main rival, Legora. Harvey’s value is in the infrastructure, the help integrating it into workflows, the ability to represent to other parties that you are using a trusted provider.

> the ability to represent to other parties that you are using a trusted provider.

so ppl trust ai if its coming from provider X ?

No they're trusting the lawyers turned solution architects that you get when you buy Harvey.

yes, esp if you're a company

it's more like you _don't_ trust ai if it comes from provider Y, or was vibe-coded by some random dev that you've never heard of

Yes.

Would you rather trust sensitive legal documents with me running a self hosted app that shares data with whichever provider openrouter chooses, or Harvey?

They trust the companies that already use Harvey or Legora for their legal work and nobody got fired for buying services from both of them.

But someone will risk getting fired for vibe-coding their own without the support, sales, infrastructure and engineering to recreate them.

Exactly. The old saying "No one got fired for buying IBM" applies here, except it's Harvey and not IBM.

This isn't that uncommon for many technologies, people and especially organizations who are procuring software look for signals that can offload perceived risk. These signals can be things as small as the formation of a company at all, or other ongoing partnerships which shows more social trust. Whether or not the company or provider is actually safe is a different conversation though.

No, they are trusting Harvy because the founders have legal/Law background and know what the customer wants. Plus the B2B contracts I'm sure they sign.

Yeah this reeks of never talking to the customers of these legaltech companies - on top of what you mentioned, it's also external integrations and downstream configurability for how those integrations behave.

If you view Harvey as Claude in a box for legal documents, then the Harvey/Legora competitor that will never have the "killer features" that larger firms receive from running Harvey/Legora.

These biglaw firms also have software engineering teams themselves. If Harvey was easy to remake as doing legal document RAG, why haven't they done it themselves?

I built for a friend's law office a RAG proof of concept. It sucked. But it was there, so I could have claimed I vibe coded an entire open source legal solution right? But it still sucked, so no (and made me believe RAG is not a useful way at all for knowledge search).

Well the firms with engineering teams are most definitely exploring doing it all in-house. There is no point in having the engineering team otherwise. I'm at a midsize firm that doesn't have that kind of staff and we're definitely going to be using something in the Harvey/Legora space, primarily because its not practical for us to roll our own.

Harvey, Mike, but no Donna!?

> Harvey

> dent

nice...

I would like to offer an alternative view here. I have about 10 years of experience as an attorney doing mostly dispute resolution (voluntarily left the bar in 2016 IIRC), and 10+ as a software dev.

It seems to me that many people believe that the primary benefit that attorneys can receive from AI is searching through droves of statutes and/or court decisions. This is certainly part of the job, especially in common law countries. But regardless of the legal system, much more substantial gains can be had elsewhere in practical day to day work. It is not often an experienced attorney needs AI to explain law to them, especially if you're working in a field you specialize in. But in almost all cases, at least in dispute resolution, an attorney would benefit a lot from an AI agent going over the matter documents to highlight stuff like this:

- possible contradictions between the opposing party's arguments now and their arguments let's say 4 years ago when they filed their statement of claim (in long running cases often parties' positions will drift over time)

- possible misalignment between the opposing party's own evidence and their arguments that depend on said evidence

- potential problems with representation, for example an important notice being signed by a person that now is a management board member, but back then did not formally have the right to represent the company

Many disputes in court are at the end of the day decided by small or not-so-small issues like these. Before AI what attorneys did was to just sit down and try to read the matter materials are thoroughly as possible. This was of course an imperfect solution. Now with AI much of this can be handed off to a machine. AI is especially good at these types of needle-in-a-haystack tasks. So in other words, in my opinion lawyers do not benefit so much from AI having an inch deep view of the whole world, but rather benefits from a mile deep view of the specific matter at hand.

Due to this I believe that law firms do not need custom LLMs, the frontier ones today are good enough for such tasks. And I also believe law firms do not need custom harnesses. Again, the ones from frontier labs fit very well already. What law firms need in practice, in my opionion, is this:

1. strict context management - for each agent session the attorney should be able to not just limit the agent to a project, but to specific files or folders inside that project, often the unit "project" is not granular engouh (think a project containing emails from the client to the attorney where they describe what "really happened" - this text should never be accessible to AI unless the attorney specifically wants it to)

2. strict agent isolation - by default a law firm should start AI agent sessions in VMs that have no internet access (other than to the AI provider and whitelisted sites) and no file system access to the machine it runs on

If these 2 requirement are effectively met, I do not see a reason why law firms would need anything legal-specific in their AI setup. Harvey and Legora provide a one-stop solution. But they also provide vendor lock-in. In a couple of years I think the natural next step for most larger law firms is self-hosted AI once prices come down to a more reasonable level. So I would encourage firms to already today try to make decisions that will likely make this transition easier in the near future.

Take all of the above with a grain of salt. I am the founder of liitmik.com (isolated AI agents for law firms), so I have a skin in the game.

No you did not. No don't have the access to jurisprudence.

I guess my first questions would be: what kind of background does the creator have in law to be able to steer this system to a good outcome? If they don't have a legal background, were legal experts and otherwise consulted to build this out? The fact that the creator is struggling with even understanding what open source is, makes it much harder to put any faith in this system.

Harvey can be pretty useful and has all sorts of tailoring done to it to provide utility to legal professionals. I don't really see any of that here. Like someone else pointed out, the main benefit of Harvey is being able to say "well we used the leading legal AI provider."

Let alone the staff to get the old fogeys at my firm using it..

I build my own take on this with https://document.bot Build for aviation legal regulations/ compliance. Highly optimized search around large PDF's, Word, etc with offline, EU and frontier LLM options.

Friend works for a fairly large law firm on the business running side. She told me how they pay a ton for Harvey and yet, they don't even have a claude for her and those not doing legal work. They think Harvey is great and have no clue that all it does is wrap around models and have safety checks like this repo does. Some of it might be yes, that there's legal work around who gets blamed, but I see mentions of Harvey and it's a joke at what the legal industry does. I see law firms as having no clue on how things work other than paying giant contracts to places that like to schmooze them because that's what they like to do as well.

Law firms are paying because the founders were former lawyers and know what the customers want and the firms trust that pedigree. As for why they pay Harvy, its about risk exposure and management. Same reason companies pay for Security appliances and vendors rather than trying to roll their own.

What risk would Harvey the company manage for them? Security appliances work, and vendors in security deal with tech that works. What is Harvey making certain in this tech?

Harvey has significantly more than "safety checks like this repo does" and if you don't realize that, you genuinely have absolutely no idea what you are talking about.

>I see law firms as having no clue on how things work other than paying giant contracts to places that like to schmooze them because that's what they like to do as well.

Okay, yeah. Going with my second option. It's really the misplaced confidence that does it for me.

This is not open source! Can we get an update to the title?

> I built the same and open sourced it

How can I *trust* you its true? Will you bet your money its __the same__ ?

27 commits, my gosh, some people are really bold in their claims.

This is incredible stuff OP Why is this being flagged?

This is literally what CodeCradt did!

Which LLM model actually built this? I looks very sloppy.

Just like how DocuSign should have shut down after thousands of other open source clones have spawned on GitHub? (DocuSign is still alive)

Also "source available" is not "open source". On top of that even releasing the source must mean that DocuSign and Harvey should be panicking right now. /s

Can someone please tell them that they should be scared?

Reply on news.ycombinator.com