# we're giving clauseo away

> we're building a new clauseo. it'll be open source and free, it'll run on the chatgpt or claude plan you already have, and every file and chat will stay on your computer.

2026-10-27 -- 15 min -- clauseo, ai, law, building

canonical: https://rohans.lol/blog/giving-clauseo-away

---
**clauseo, before and after.** new clauseo isn't built yet. this is what it will be.

| | old clauseo | new clauseo, once it ships |
|---|---|---|
| what lawyers pay clauseo | about ₹25,000 a month each, india only | nothing |
| how much lawyers use it | four or five tasks a month | no limit from us |
| the code | closed | open source |
| where clauseo runs | our cloud. your laptop was a window onto it | an app on your computer |
| your files and chats | on our cloud | on your computer. nothing of yours will be on our servers |
| the ai | we paid the ai company, then billed you | the chatgpt or claude plan you already pay for |

old clauseo held lawyers back twice. it cost about ₹25,000 a month for four or five tasks, so they rationed it. every task meant handing us privileged documents. new clauseo will be free and open source. it'll run on your computer, on the chatgpt or claude plan you already pay for. your files won't ever reach us. you'll decide what goes to openai or anthropic.

the last two months have been shit. clauseo stopped growing, i fell pretty low, and most weeks i was doing the maths on how long we could last.

at first i thought the problem was me. i felt guilty that i wasn't selling hard enough, and then i started doubting whether i could sell at all, which is fucked, because i've sold before and i'm not bad at it. so i did the most me thing possible and tried to build my way out. agents to find lawyers and write my outreach, then an app with cards that made me message three people a day. that's how deep the hole was.

then i noticed every conversation ended on the same two questions. this is too expensive. and where do my documents go? i didn't have a real answer to either. my answer was always "we'll build your firm its own deployment", which meant a long sales cycle and nothing anyone could try today.

i tried free credits. ₹3,000, ₹5,000, whatever it took. people tried it, loved it, told me nothing else came close, and never came back. they couldn't afford to use it for real, and they couldn't put their real documents in it.

i've never liked selling. the best salespeople go find the easiest product to sell, because a product that's hard to sell is a shitty product. and clauseo was a shitty product. not because the agent was bad. the agent was fucking great. it was shitty because of its shape. a consumer app, for a profession that can't do consumer apps.

the weird part is that by one number, we were doing great.

our paying users average about ₹25,000 a month each. that's more than $250, for maybe four or five tasks. most ai products would kill for that number.

four or five tasks a month. they were rationing it. clauseo is expensive, so lawyers saved it for the urgent stuff with the tightest deadlines, and before every question they asked themselves if it was worth clauseo. i hate that. i don't want anyone thinking twice before using it. i want them to throw everything at it, experiment, and use it for the boring stuff too.

same two problems. the price, and handing a startup privileged documents.

so i'm killing that number on purpose, and the old clauseo goes with it.

we're building a new clauseo from scratch. it'll be open source and free, and it'll run on the chatgpt or claude plan you already pay for. every file and every chat will stay on your computer. we'll never see any of it.

it's the biggest change we've ever made, and it throws out most of what i've written here about how clauseo works. and for the first time in months, i'm excited again.

## mahindra never sold me diesel

clauseo has always run on a simple deal. you pay us for tokens, which is just ai usage, we pay anthropic or openai, and we keep a cut. i wrote [a whole post](/blog/clauseo-not-a-chatbot) defending it.

the problem is that we were competing with the labs at selling their own product. openai and anthropic sell $20 and $200 plans that are worth way more than that in raw usage. sam altman [said openai loses money](https://techcrunch.com/2025/01/05/openai-is-losing-money-on-its-pricey-chatgpt-pro-plan-ceo-sam-altman-says) on the $200 one. they can eat that. it's their fuel. we can't. so a lawyer looks at their $20 chatgpt bill, then at ₹8,331 for [one big matter on clauseo](https://www.clauseo.chat/demo/34506b4c-6301-4638-a10f-8ff1c06a7e47), and the conversation is over before it starts.

the labs are fuel companies. we make cars. my thar, jessica, didn't come with a lifetime supply of diesel from mahindra, and nobody thinks that's strange. you buy the car, you fill it up at whatever pump you like.

ai startups decided to sell the car and the fuel as one bundle. go look at how that's going for [jasper](https://www.maginative.com/article/jasper-cuts-internal-valuation-as-ai-growth-slows), [windsurf](https://techcrunch.com/2025/07/11/windsurfs-ceo-goes-to-google-openais-acquisition-falls-apart/), [cursor](https://openai.com/index/our-decision-on-cursor-following-its-acquisition-by-spacex) and [harvey](https://thenextweb.com/news/model-costs-startups-open-weights). the lucky ones get bought by a fuel company. legal ai doesn't even get that exit, because the only thing a lab would want from us is our users' data. that was never ours to sell. if we ever tried, every lawyer on clauseo would sue us, and they'd be right.

**the car and the fuel, drawn out.**

| | old clauseo, selling fuel | new clauseo, building cars |
|---|---|---|
| fuel money paid to clauseo | ₹8,331 for one big matter | nothing |
| where that money goes | most of it went to the lab as the token bill, clauseo kept a thin cut | it won't come through us |
| the fuel | resold from openai or anthropic | your own chatgpt or claude plan, at whatever pump you like |
| what it's up against | the lab's own $20 plan, with far more usage in it | nothing, it'll run on that plan |

**four companies that sold the car and the fuel together**

| | what happened |
|---|---|
| jasper | $1.5b, then chatgpt launched |
| windsurf | anthropic cut off its claude access with under 5 days notice |
| cursor | ripped up its $20 plan. spacex bought it, and now openai is cutting it off |
| harvey | gross margin went from about 50% to about minus 50% until it post-trained its own model, tenet |

the lucky ones get bought by a fuel company. so we're cutting the fuel line. new clauseo will be just the car, and you'll fill it up wherever you like, on your own plan.

i have a lot more to say about this one. it might get its own post.

## a very expensive way to fake a laptop

we built our own engine too. the industry calls it a harness. it's the program that turns a model into something that does work. it reads files, runs code, searches, spawns helpers, checks itself, and loops until the job is done.

ours ran in the cloud, because we wanted clauseo on any device. so every conversation needed a computer, and we built one out of parts, every single time. more than thirty thousand lines of it. i [wrote it up](/blog/how-to-build-a-harness) eight weeks ago, very proudly.

**the fake laptop, drawn out.**

| | old clauseo | new clauseo | 30 at once |
|---|---|---|---|
| where the computer was | built far away, out of parts, for every conversation | the laptop already on your desk | a row of machines in the firm's own cloud |
| the parts | a rented server, a frozen computer, your files copied over from our storage, wiring to join them | none | the firm's servers behind its fence, with every action logged |
| what each message did | woke that machine up for one turn and sent the answer the long way back | will go to your chatgpt or claude plan, and the answer will come straight back | will run on the firm's machine, with the model under the firm's own zero data retention terms |
| after the conversation | the computer got thrown away | nothing to throw away | the machines will stay on and the files will stay with the firm |
| code to make it work | more than 30,000 lines, just to fake a laptop | none of ours | the cloud setup from the harness post, on the firm's machines |

your laptop was only ever a window. the computer we kept building was sitting on your desk the whole time.

once a lawyer runs 10 or 30 tasks at once, one laptop will choke. cloud machines will make sense again, as long as they're the firm's and the files stay with the firm. clauseo won't hold them.

the ordinary computer was sitting on your desk the whole fucking time.

and because every message had to wake up a machine on the other side of the planet, it was slow too.

i still stand by that post. if you're building managed cloud agents, that's how you build them. and once a lawyer runs ten or thirty tasks at once, a laptop chokes and cloud machines make sense again. it's just the wrong build for the lawyers we have today. they run a few matters at a time, and unlike code, their files are privileged. cloud agents are fine. cloud agents on our machines, holding your client's documents, are the problem.

and the labs want to own the engine anyway. claude code, codex and cowork are all turning into the thing clauseo was built around. the most tech obsessed lawyers i know already skip legal ai and run claude code with a folder of skills. in february anthropic shipped a legal plugin for cowork and [legal data stocks dropped double digits](https://www.morningstar.com/stocks/reuters-relx-wolters-stocks-crushed-after-anthropic-debuts-claude-legal-plug-in).

i'm not precious about it either. clauseo v2 was already a rewrite from scratch, and i rewrite everything about every six months. what i build next is always better.

## trust me bro, enterprise edition

cost was bad. trust was worse. why would a lawyer put privileged documents into clauseo at all?

to run in the cloud, we had to hold your files on our side. we never wanted them. i never asked you to trust me. i don't want you to. i have zero interest in reading your client's documents. why would i? we held them because the architecture made us.

the big names have the same problem with better marketing. harvey and legora are cloud products. your files live on their servers. they put real work into protecting them, with encryption, certifications and retention promises, but the files are still there, and the protection is a policy you have to trust. it's trust me bro with excellent paperwork. and apart from [mike](https://github.com/Open-Legal-Products/mike), an open source project from earlier this year, nearly every legal ai works the same way. you upload first and trust later.

lawyers don't trust startups, fair enough. people trust google with their email and wouldn't trust me with it, and that makes sense. but they don't want to trust anyone. they don't want privileged material sitting in six places it could leak from. and whether a lawyer is okay sending documents to openai at all is their call. it was never ours to make. if you're building legal ai, a product that never even touches your users' data is the dream.

i get it, because i'm the paranoid one. for three years i haven't touched chatgpt, and i've never used claude code. i use my own harness, on api keys, with everything stored on my own machines. none of my ai usage has ever trained a model. i just don't see why my stuff should live on somebody else's computer. turns out that's exactly what lawyers wanted. i had the answer the whole time and never looked at it.

## the car (free, which still hurts)

the new clauseo will be an open source desktop app for lawyers, on mac and windows. open source means the code is public. anyone can read it, run it or change it.

under the hood it'll drive the agents the labs already make, claude code and codex to start, through [acp](https://zed.dev/blog/bring-your-own-agent-to-zed). acp, the agent client protocol, is a standard plug zed made last year that lets an app drive a coding agent. the agent and the plan will be yours. clauseo will be the interface on top, made for lawyers, not for coders.

why would a lawyer's app drive a coding agent? because models got good at legal work by getting good at code. clauseo already doesn't research the way a human does, one search at a time. it writes scripts that run hundreds of searches and filter what comes back. no legal expertise taught a model that. coding did.

here's an early design mockup, tap it for the full size. the matter is made up, and yes, that's a debug box in the corner.

![early design mockup of the new clauseo, with a made-up cheque case, rathi steels v. northgate infra. on the left, matters sorted by the next hearing. in the middle, clauseo's answer on whether the complaint against an independent director can be quashed, and the other side's best case. on the right, the matter's papers numbered by paper book page and marked supports or against, and citations marked good law. a debug box sits in the top right corner.](https://rohans.lol/blog/giving-clauseo-away/new-clauseo-design.png)

## the engine was never the personality

a lot of what made old clauseo good was being on the bleeding edge of harnesses. that edge stays. i build harnesses myself. [cmx2](https://github.com/clauseo-ai/cmx2), the one i run my own agents on, is open source and has nothing to do with clauseo. the new clauseo will ship with claude code and codex tuned the way i tune my own, so you never have to think about any of this.

everything that isn't the engine is still clauseo. the legal tools and the data sources. the citation checks. which model does what, and when to hand work to a sub-agent. the prompting, which needs less of me every time a new model ships. even the way it talks to you.

there's more under the hood. every matter will be its own git repo, which is a nerdy way of saying every change gets saved with a history you can roll back. there's a branch with about 120 legal skills that never shipped, and an interactive viewer that never shipped either. those come after launch. getting the new clauseo out comes first.

none of that should be your problem, though. you download clauseo, sign in with your subscription, and go. that's it.

which means:

- **it'll be free.** no tokens to buy from us, ever.
- **you'll bring your own ai.** you'll sign in with chatgpt or claude. more providers as their terms allow.
- **everything will live on your computer.** every document, every chat, every matter, every draft. none of it will touch our servers.
- **it'll be open source.** if you don't believe a word of this post, read the code once it's up.

that third one is the whole point. flip between old clauseo and new.

**where your stuff lives.** new clauseo isn't built yet. the last two columns are what it will be.

| | old clauseo | new clauseo, once it ships | at a firm |
|---|---|---|---|
| your documents | your laptop, our storage, and a copy in a rented computer for every chat | your computer | the firm's own cloud |
| your chats and matters | our database | your computer | the firm's own cloud |
| the ai company you picked | sees what each chat sends it | sees what each chat sends it | sees what each chat sends it |
| clauseo's servers | held your files and chats | nothing of yours | we'll only get in when the firm lets us |
| places your stuff lives | 5 | 2 | 2 |

old clauseo kept your files and chats in five places. your laptop, our storage, our database, a rented computer for every chat, and the ai company. new clauseo will keep every document, chat, matter and draft on your computer. it'll be an app on it and nothing more. the only thing that'll leave is what each chat sends to the ai company you pick. at a firm it'll be the same, inside the firm's own cloud. the fence will log every action, and we'll only get in when the firm lets us. the only line out will go to the ai company the firm picks.

your documents will take one line out of your computer, to the ai company you picked. there won't be a clauseo server holding your stuff, because there won't be a clauseo server for your stuff at all.

and we're not an india-only company anymore. india comes first, because that's where our legal layer lives today, and a us configuration comes next.

## but i use clauseo on my phone

i know. a lot of you do, mostly between hearings. so how does your phone reach an app on your desk if there's no server in the middle?

it'll talk to your computer directly. the plan is iroh, an open source library for peer-to-peer connections. you'll pair once, and then your phone and your computer will open an encrypted line, just the two of them. on a locked down office network the traffic will bounce through a relay that can't open what it's carrying. your phone will be a window into your computer, not a second place your files live. it'll come after the desktop app.

the catch is that your computer will have to be awake. firms will get machines in their own cloud that don't sleep.

**how your phone will reach your computer, drawn out.**

| | what happens | where your files are |
|---|---|---|
| direct | pair once, then phone and computer will open an encrypted line, just the two of them | your computer. the phone will be a window into it |
| locked down office network | the direct line gets blocked, so the traffic will bounce through a relay. it'll be sealed end to end, and the relay will pass it on without being able to open it | your computer |
| laptop asleep | close the lid and the phone won't reach anything until you open it again | your computer |
| at a firm | the phone will talk to a machine in the firm's own fenced cloud. it won't sleep | the firm's own cloud |

there won't be a clauseo server in the middle. the phone app will come after the desktop one.

## no lock-in, including us

this is the part i'm most excited about, and it's for firms.

right now a firm that wants legal ai signs a big per-seat contract and hands its documents to a startup. with the new clauseo, nobody will have to trust us. the code will be auditable and we'll never hold the data. for firms, starting with small and mid-sized ones, we'll set it up properly: agents running on machines in the firm's own cloud instead of on a hundred associates' laptops, ai usage under zero data retention terms the firm signs with its cloud provider directly, storage the firm controls, every action logged. that's the cloud setup from my harness post coming back, on the firm's machines this time. no cloud? we'll set one up in the firm's name. either way the machines and the data will be the firm's, and we'll only get in when the firm lets us in.

and no lock-in. not with the labs, because you'll be able to swap claude for chatgpt or whatever comes next in an afternoon. not with us, because it'll be open source, and if you get sick of us you can hire someone else to keep building on it. agents [made custom software cheap](/blog/let-the-users-build), so a document review or due diligence pipeline on top of clauseo will be a week of work, not a procurement cycle.

i've watched enough companies squeeze their customers. it always ends badly, maybe not now, but eventually. that's not how i work.

**the plug and the swap, drawn out.**

clauseo will be the legal layer, with legal tools, case law, prompts, matters and citation checks, and the word file at the end. it'll sit on acp, the standard plug, and the engine underneath will be yours.

| engine | runs on | what changes in clauseo |
|---|---|---|
| claude code | your claude plan | nothing |
| codex | your chatgpt plan | nothing |
| more | as their terms allow | nothing |

you'll be able to switch the agent underneath in an afternoon and nothing in clauseo will change. it'll be open source too, so you can take your own copy and hire someone else to keep building on it.

## for the vcs reading this with a spreadsheet open

yes, this kills our only revenue line. on purpose.

clauseo is becoming an open core company. the core, the app every lawyer will use, will be free and open source. firms will pay for the ring around it: deployment on their own cloud, admin and audit, wiring into their document systems, custom pipelines, and someone to call when something breaks.

**the business, drawn out.**

| | the old books | open core |
|---|---|---|
| what lawyers pay us | about ₹25,000 a month each, for four or five tasks | nothing, the app will be free and open source |
| who pays us | lawyers | firms, for the ring around the core |
| what firms pay for | nothing, there was no firm product | deployment on their own cloud, admin and audit, wiring into their document systems, custom pipelines, someone to call |
| the token bill | ours, resold fuel | off our books, you'll bring your own chatgpt or claude plan |
| client documents | ours to hold | off our books, they'll stay on your computer or your firm's cloud |
| cash fronted to the labs | ours | gone |
| what we sell | resold tokens | software and setup |

in the old books lawyers paid us about ₹25,000 a month each, and most of it went straight back out to openai and anthropic as the token bill. we also sat on their client documents. the margin was thin. now the app will be free for every lawyer. firms will pay us to set it up on their own cloud and keep it running. the token bill and the client documents will leave our side.

look at what leaves the books. no token bill in our costs, so we'll sell software and setup, not resold fuel. no client documents on our side, so nothing for us to leak. and the marketing budget we never had stops mattering, because the product will be free and will work outside india.

and yes, open source means anyone could sell that setup. firms will pay the people who built it and know it best, same as they always have with open source. and if harvey or legora want to copy the code, go ahead. they'd have to run it on their own servers, holding your documents, and that's the exact problem we're getting rid of.

there's a second thing we'll sell. case law search will be a paid add-on, and it'll be closed source. it's an mcp server, the standard plug agents use for tools, so claude, chatgpt or any other agent can use it, not just clauseo. the app stays free either way.

the plan to embed our people inside firms, the way palantir does, survives. the part where those deployments fed our data doesn't. the firm's work will stay the firm's.

## the asterisks

lawyers read the fine print, so here it is.

- **your ai company will still see your documents.** we'll be out of the loop. the lab won't. in february a federal judge in new york [ruled that documents a defendant made with consumer claude weren't privileged](https://www.harrisbeachmurtha.com/insights/in-a-first-court-finds-ai-generated-documents-not-protected-by-attorney-client-privilege/), partly because the consumer terms allowed training and disclosure. so for client work, use a business or enterprise plan that doesn't train on your data and comes with confidentiality terms. yes, the paranoid guy is telling you to bring your chatgpt account. the difference is you pick the one company that sees your files. with most legal ai you get two, and you didn't pick the second.
- **a $20 plan will get you started, not far.** it'll be enough to try clauseo properly, which is more than $20 ever bought you from us. heavy matters want the bigger plans, and you'll hit their limits a lot later than you'd hit our prices.
- **the labs keep changing the rules.** which plans work in which apps has changed several times this year. today claude code and codex both take your own plan. if a lab ever shuts its plans out of other apps, you'll plug in an api key instead, or switch to another lab. if that happens, we'll say so.
- **case law search** will be the paid add-on. your search terms will reach it. your documents won't.
- **phone notifications** are the one piece we haven't worked out yet, because phones only wake up through apple's and google's push servers. when we build them, we'll say exactly what goes through.
- **we don't know our number yet.** revenue was easy to count. open source isn't. we haven't settled on how we'll measure whether this works. the plan is opt-in telemetry that counts how much clauseo gets used and nothing about what's in it.
- **it's not out yet.** we're building it right now. if you use clauseo today, nothing switches off tomorrow, and you'll hear from me before anything changes.

## yes, i know how this looks

people are going to think i've lost it. paying users averaging more than $250 a month, and i'm giving the thing away.

and yeah, it hurts. i've shut down things i built before, and it still hits every time. this one's the scariest. it's the only revenue we had, and i'm switching it off before anything replaces it. until serious firms run clauseo, there won't be any. the innovator's dilemma says companies never kill the thing paying them. i'm killing it myself, before someone else does.

but give me the choice between revenue and lawyers loving clauseo, and i'll pick the lawyers every time. revenue is something you optimise much, much later. not now.

the billing system goes with it. the wallet, the top-ups, the meter counting every token a lawyer used. it's my least favourite thing i've ever built. i hate putting a price on something i love. in march i spent an evening on it, then put on nirvana unplugged and [wrote a sad post about kurt cobain](/blog/nirvana-the-tshirt-company) instead. now i get to delete the fucking thing, and i can't stop laughing.

this is the only version where a law student, a solo practitioner, someone in chambers, an associate at a big firm and a general counsel will all be able to use the best legal agent we know how to build, without paying us and without trusting us. all you'll need is a chatgpt or claude account.

my open source career so far is a handful of fixes for bugs i spotted in projects i use. i've never run one that people depend on. now the thing i'm proudest of building is about to become one. it'll be new to me, and i'm doing it anyway. and if clauseo the company dies, the code won't. it'll stay out there for anyone to use, instead of rotting in a private repo.

there's one more open source thing landing before clauseo does. it's called [stetdoc](https://github.com/clauseo-ai/stetdoc), and it's getting its finishing touches. agents that edit word files by going through markdown lose the formatting, the tracked changes and the comments. stetdoc lets an agent edit a word file and hand it back with real tracked changes, every edit a revision with an author and a date. it's how clauseo will edit your word files too.

goodbye, old clauseo. you were the best thing i'd ever built.

i hope the next one outlives you.
