Own The Platform — Source Licence
Your firm should build software around its own process, not bend its process around someone else's software. But we have watched too many firms lay that foundation themselves — and regret it eighteen months later. Obe is the other option: start at ninety percent, and build the last mile.
I. The Foundation Problem
A firm decides to build. The first six months look fine — a form, a database, a folder of PDFs, a script that renames them. Nobody is thinking about how a citation survives a re-extraction, or what happens when two reviewers correct the same field, or how you prove eighteen months later which schema version produced a number that ended up in a demand letter. Those are data-engineering decisions, and they are being made by accident, early, by whoever is closest to the keyboard.
By the time the firm needs those answers, the foundation cannot give them. Not because anyone was careless — because a foundation that supports document intelligence has to be designed for it from the first table. Rebuilding at that point costs more than the original build, and it costs it during the year the caseload finally arrived.
II. Build From Ninety, Not From Zero
Nothing about extraction state machines, citation provenance, confidence calibration, or tenant isolation is where a law firm wins. Those are table stakes that take years to get right and are invisible when they work. What differentiates your firm is the last mile — your case types, your screening thresholds, your review process, your theory of the claim. Obe hands you the ninety percent that is expensive, undifferentiated, and already in production, and leaves the ten percent that is actually yours.
This is the more conservative path, not the more adventurous one. Buying a foundation that already runs is how you take risk off the table, provided the foundation is real.
III. What The Ninety Percent Actually Is
There is a genre of legal-tech repository that is a form, a prompt, and a hope. This is not that. It is an AI-first platform with the unglamorous parts already built — the parts a demo never shows you and a production caseload finds within a week.
Staged state machine with retries, backoff, dead-lettering, and monitoring. Per-document-type schemas, versioned and fingerprinted.
Every extracted field carries a citation to its page and bounding box, a confidence, and a support status. Verification rules enforce citation coverage.
Confidence thresholds recomputed per pipeline from what reviewers actually corrected — not a number someone guessed once.
Organisation scoping derived from the credential, never from a request parameter. Cross-tenant reads are rejected and audited.
Your case and intake data reachable from Claude Cowork or any MCP client, scoped by key, advisory by contract.
Per-case-type legal issue detection, theory scoring, limitation clocks, leverage and damages analysis, with the evidence behind each signal.
IV. Two Licences
A perpetual source licence. No seat count, no renewal, no clause that lets the price move once you depend on it. You run it on your own infrastructure and you extend it with your developers, or with ours.
Firm Licence
$70,000
The per-case-type screening system
Full Source
$140,000
Case management, arbitration intelligence, and intake
V. What You Should Know Before You Buy
A six-figure licence deserves the caveats up front, not on page forty of a data-room document. Here is what owning this actually entails.
VI. Questions Firms Ask
What does "own forever" actually mean?
A perpetual licence to the source, for your firm. No seat count, no annual renewal, no mechanism for us to reprice you once your practice depends on it. You deploy it on your infrastructure and you modify it freely.
What does it cost to run once we own it?
Hosting is around $90 a month in typical firm use. On top of that you pay your own AI-provider and document-processing usage, at whatever rate you negotiate with the providers you choose. There is no revenue share and no per-case fee.
Are we locked into your AI vendors?
No, and that is the point of buying rather than renting. The provider registry is open and model selection is configuration. If a better or cheaper model appears next quarter, you switch to it without asking us.
Does it work for our case types, or only solar fraud?
The engine is case-type agnostic — document types, schemas, and screening packs are all definable. The pipelines that ship are built around consumer-finance and solar-fraud documents, because those are the cases we fight. Yours are defined in onboarding, and you can define more forever without us.
Who builds the last mile?
Your developers, ours, or a mix. Owning the source means you are never blocked on our roadmap. Firms that want it done for them engage us; firms with an engineering team take the code and go.
Why is this less risky than building it ourselves?
Because the expensive, invisible decisions are already made and already running against real caseloads — extraction state machines, citation provenance, confidence calibration, tenant isolation, audit trails. Those are the decisions a from-scratch build gets wrong quietly and discovers late. Starting at ninety percent moves the risk from architecture, where it is fatal, to configuration, where it is cheap.
Elsewhere For Law Firms
VII. Talk To Us
A licence conversation starts with what your firm actually fights and what it already runs. Book a call, or write to us.
Weighing build against buy? Request the architecture overview — what the ninety percent contains, before you get on a call.