Skip to content

Company

Law firms run on what nobody has had time to read.

Caseforth is building the system that reads it: every document on every file, every day, with agents that do the work it calls for and people who decide what leaves the firm.

For the first time, a whole book can be read first hand.

01

Models can read a whole file

Not by pasting it into a chat. An extraction pipeline built for models turns every document into clean text with its pages, and agents work through the entire file, knowing what they have and have not read.

02

Agents can do the work

Reading is no longer the end of it. Agents can decide what a matter needs and prepare it, and a second model can check the call.

03

The cost has collapsed

Flat-rate subscriptions put a whole book within reach of a firm, where metered reading at this scale was out of reach.

04

Risk is the price of not reading

Missed limitation dates, unanswered clients, and unpaid settlements are what firms insure against. A firm that reads every file carries less of that risk.

How it was built.

Ten months, over 7,000 commits, and one idea held throughout: the truth about a case is in its file, so read the file.

  1. November 2025

    A drafting add-in

    Letters generated inside Word, in the document a fee earner already had open.

  2. Early 2026

    From templates to the file

    Drafting moved onto the evidence: facts read from the documents, contradictions caught, case assessments written from what the file actually says.

  3. Spring 2026

    Local and in-house

    Off rented cloud services, onto local, in-house machines and document stores, where the data stays.

  4. June 2026

    Matterframe

    The model of legal work was separated from its first practice, so what is learned for one kind of claim serves every other.

  5. Summer 2026

    Specialist readers

    Bank statements reconciled to the bank's own balances, medical records read for capacity, audio and video evidence transcribed.

  6. September 2026

    Caseforth

    Caseforth Ltd was formed and Desk was built. Every document on the book now has one reading, every obligation sits in one register, and agents do the work before anyone asks.

All of it now sits in one place: a unified model of legal superintelligence, with agents that know the work is needed before anyone asks for it.

Leadership.

PW

Paul Wood FRSA

Founder and Chief Executive

Paul Wood founded Caseforth to build a unified model of legal superintelligence: one that reads every file a firm holds, knows the work each matter needs, and does it before anyone asks. His research into how software sees what is really there, rather than what it assumes, runs through the company's approach to reading a firm's files.

Security research

Accepted into Anthropic's Cyber Verification Program and OpenAI's Daybreak, which give security researchers access to their models for vulnerability research.

confusable-vision

The first font-by-font dataset of Unicode characters that pass for one another, measured from 322 fonts. Used in Mozilla's add-on name checks, disarm, and SilverSpeak.

namespace-guard

Checks whether a name is really free, including against lookalikes. Listed in Anthropic's Claude Code plugin directory.