Makor gathers the material of a matter, filters it, reads it, cross-references it and verifies it. What comes out is a single factual record: facts, people, a chronology, contradictions and gaps. Every item in it leads back to the document it came from.
The lawyer builds the strategy. Makor makes sure the factual picture it rests on is complete, organised and open to inspection.
The problem
A single commercial dispute rests on thousands of documents, message threads, contracts, spreadsheets, invoices and filings, built up over years and scattered across mailboxes, shared drives and scanned binders. No team can read all of it, hold all of it in mind and cross-reference all of it against the court's timetable.
So the factual layer the strategy rests on is built by hand, usually by the junior team, in a spreadsheet and a Word file. The partner making the most consequential decisions in the matter relies on work whose completeness rests on people's memory and on the time they had.
Sources
Makor ingests mailboxes, shared drives, folders dragged in as they are, WhatsApp exports, old scans and court filings. It spots duplicates and versions, links every item to what it proves, and reads what was not meant to be read: hidden spreadsheet rows, deleted comments and version history.
Collection is scoped to the matter, not to a person. A colleague's mailbox is never taken wholesale: it has to be narrowed by keywords, or by intersecting it with the opposing side's participants, and the system itself enforces that.
What Makor builds
Makor does not produce another folder of files. It builds one layer that holds the matter: every fact, who said it, when, which document it is written in, and where its evidence stands. The chronology, the evidence matrix, the contradictions, the gaps and the pleading map are all views of that layer, not separate documents each needing upkeep of its own.
New material does not simply land in a folder. It is checked against what the matter already knows: a claim that had no evidence finds its match, an event on the timeline gains a supporting document. Every match goes to a person for approval, and only then does it enter the record.
Where Makor comes in
Teams change and the material keeps piling up. Makor keeps the factual ground in one place from day one, and feeds it into the tools the firm manages, writes and files in.
Who it is for
The engine, the quote verification and the security are identical in all three. What changes is the volume of work, the setup, and who administers the system day to day.
A litigation department running evidence-heavy matters in parallel, with teams that change over the years.
Exactly the same system. Setting up means uploading the material, and the firm carries on working as it does.
The company's own material, gathered around a single dispute: mailboxes, shared drives, contracts and operational correspondence.
A small firm gets the same system, with a number of users and open matters that fits its size. Quote verification, the matter wall, the audit log and how material is retained are identical across all three.
Deployment and security
Two options, the same system, the same permissions and the same audit log. The difference is who operates the infrastructure.
On the firm's infrastructure
A dedicated private server we operate
Exactly what leaves, and when: the matter's material stays out of any Makor cloud. When an analysis runs, the relevant passages are sent to the model provider from the server running the system, under the firm's own account and key and according to what you approved. Intake, filtering, scan decoding, Hebrew search and quote verification all run on the server itself.
How it works
In
Processing
Out
Local processing and cost
Makor does not send every operation to an external model. Scan OCR, metadata extraction and image analysis run inside the private environment where the system is installed, and anything triage rejected does not proceed to analysis. Less material leaves, the cost falls, and you know exactly what was sent.
Hebrew and English decoding runs on the server that runs Makor, on the machine's own account, and the document stays there. Every page gets a confidence level, and only weak pages can be sent for advanced paid decoding, if you choose that.
A small vision model in the same environment writes a short description of each image, and the file itself yields the capture date, the device and the location if one was stored. The description is a finding aid; the evidence stays the document itself.
Material marked irrelevant does not proceed to model analysis and incurs no model cost. Filtering happens before the expensive stage, not after it. The material itself stays searchable, so nothing set aside during triage ever becomes unreachable.
Spend is metered per matter, shown as it happens and stopped at a ceiling you set. An expensive model is used only where it genuinely adds something; whatever can be done locally, or done deterministically, stays there.
The metering is there so you know, and the ceiling is there so a question can be asked freely. The overwhelming majority of the work runs locally, on the machine's own account.
Trust and verification
A model can point at where the material is, and a quote is always pulled from the document. Deterministic code pulls the original document and checks the quote against it, character by character, and only then does the quote get a status. A quote that is not found as written is not quietly dropped: it is flagged, and the fact resting on it is downgraded and reaches the lawyer marked.
A statement that rests on inference is marked as inference and shown as one, so the line between what the material says and what was learned from it stays visible.
A scanned document is decoded, and the text produced becomes that page's text layer. A quote from a scan is checked against the text layer of the page it came from, and the page number is kept so the scan itself can be opened and read. A description the system writes for an image, or metadata read out of a file, serve to find material: verification rests on the document's text layer alone. A layer created to help with search stays an aid to search, and the evidence is the document itself. That distinction is written in code, not in a procedure.
The quote appears in the document exactly as shown. This is the highest verification level the system gives.
The meaning holds but the wording differs from the original. The item is marked, and stays labelled as a paraphrase.
The quote is not in the source. The fact is downgraded automatically and reaches the lawyer flagged in red.
An exact quote, in the original, where it was cited.
Several consistent sources point the same way.
There is evidence, but a link in the chain is missing.
Inferred from the material and labelled as inferred. Never quietly.
Sources conflict. Flagged for the lawyer's attention.
Inside the workspace
| Fact | Source | Who | Date | Evidence grade |
|---|---|---|---|---|
| Delivery date set at 30 June 2024 | EX-7 §4.2 | Both parties | 12.1.23 | Proven |
| A nine-week slip was reported internally | EX-142 | Site manager | 3.8.23 | Well supported |
| A supplier failure caused the slip | EX-142, EX-155 | Contractor | 8/23 | Partial |
| The developer knew of the slip before the September report | – | – | – | Inference |
| The schedule was on track in September 2023 | EX-198 | Contractor | 15.9.23 | Contradicted |
Who knew about the delay before the September progress report?
The site manager reported a nine-week slip on 3 August 2023, six weeks before the progress report. It was sent in a WhatsApp group that included the developer's own project manager.
Did the developer approve an extension of the delivery date in writing?
There is no evidence of that in this matter. No document was found in which the developer approves an extension. Searched: extension of time, schedule approval, deferred delivery, additional days. Zero verified hits.
Makor answers only from what is in the matter. Where there is no evidence it says so, and shows what it searched. Every quote in an answer was checked against its document before it was shown.
| Paragraph | Claimant's allegation | For | Against | Status |
|---|---|---|---|---|
| §12 | Delivery was eleven months late with no prior notice | F-0142, F-0155 | F-0201 | Mixed |
| §14 | The defendant never reported any schedule delay | – | F-0142 | Contradicted |
| §19 | The claimant paid the full sum due for milestone 4 | – | – | No evidence |
| §23 | A written notice was sent on 2 February 2024 | F-0311 | – | Supported |
The pleading goes in, and every paragraph comes back mapped against the material: what supports it, what contradicts it, and where there is nothing at all. The map is raw material for the lawyer's decision, not a legal position.
Everything in these views belongs to a fictional matter built for illustration. This site contains no client material and no real names.
Capabilities
Every quote is checked character by character against the document's text layer. Anything not found as written is flagged, and the fact resting on it is downgraded.
Scattered documents become one table of facts: what is asserted, who said it, when, from which document, and at what evidence grade.
The full chronology of the matter, with every event linked to the evidence it was drawn from.
Ask in plain language, get an answer drawn only from the material, with a link to every source. Where the matter holds no evidence, the answer says so rather than guessing.
Inconsistencies between documents, statements and pleadings, with both quotes side by side. And alongside them what is absent: the evidence you would expect this matter to contain and it does not.
A system that shows you only what you hoped to find is not a litigation tool. Material that contradicts or weakens your client's account is not an exception here, it is required output: flagged and surfaced, rather than waiting at the bottom of the pile until cross-examination.
A pleading goes in, and each paragraph comes out mapped against the material: what supports it, what contradicts it, and where there is no evidence at all. The legal conclusion stays with you.
The cast of the matter is built from the correspondence itself: addresses, nicknames, WhatsApp display names and spelling variants. Duplicate identities are proposed for merging.
A file entering the matter is checked immediately against what the matter already knows: a claim that was waiting for evidence, or a date already on the timeline. The match goes to a person to approve, and the quote is verified again at the moment of approval.
An ordered research file, and a numbered document list towards disclosure, with items suspected of privilege flagged for review. All of it a working draft for a lawyer to check and sign.
The list of documents no issue in the matter has touched. The analysis also shows what stayed outside the pleading, so the decision is made on the full picture.
A witness, an employee or a client can be sent a secure link that lets them connect the relevant mailbox or drive themselves, with the password staying with them. Consent is recorded, the state of the connection is visible at any moment, and every grant can be revoked.
What you get
The deliverables come out in chapters. A chapter stands on its own, at a size a writing tool reads comfortably, and inside it are the facts with the quote and the document behind them.
A chapter per theme: what happened, who said it, from which document, and at what level of evidence. Every chapter is edited in Hebrew and marked as a draft for a lawyer to check.
Facts and sources, the chronology and the pleading map, as tables you can filter, sort and work from directly against the material.
Every paragraph of the pleading against the material in the matter: what supports it, what contradicts it and where a gap sits. That is the worklist the next paragraph is written from.
Makor opens as a read-only interface, so the assistant or the writing tool you work in pulls a fact with its quote directly. Opening it is a firm setting, the token belongs to a named lawyer, and every read is written to the audit log.
The drafting stays with you and with the tool you write in. Makor supplies the material it rests on, ordered, verified and ready to feed in.
Recurring questions
You connect the sources: mailboxes, drives, folders and WhatsApp exports. A lawyer approves the settings of the matter, and only after that approval does the analysis begin to run.
Only the passages that passed filtering, from the server running Makor, under the firm's own account and key and according to what an attorney approved. Every call is written to the audit log, and a firm that chooses it runs a whole matter on its own server.
Cost is metered and shown in real time, every matter has a ceiling you set, and most of the work runs locally, on the machine's own account. The metering is there so you know, and so a question can be asked freely.
Yes, and that is the default posture. A dedicated private server that we operate is also available, and infrastructure in Israel can be chosen.
The installation keeps working, the licence is checked locally, and there is source-code escrow and a full operations runbook. The detail is on the Security and trust.
Chapters, tables and a read-only connection to the tool you write in, as set out in What you get.
Another question: write to us and we will answer in writing before the meeting. You can also read how an introductory meeting runs and see How the price is built.
Where it fits
The boundary
Makor brings the material to a state where a decision can be made. The decision, the drafting and the strategy stay with the lawyer and with the tool they write in.
Makor is responsible for
The lawyer is responsible for
Book a demo
Thirty minutes on a fictional matter built for the purpose. You do not need to bring real material to see how it works. Tell us about the kind of case weighing on you now, and we will show you what it looks like once Makor has been through it.
No client material is used in demonstrations.
Book a meeting
A 30-minute introductory call, straight into our calendar. Pick a free slot, leave your details, and the invitation arrives in your inbox.
Times are shown in Israel time.
Loading available times…