Darwin Axiom

Axiom is the Darwin Suite's product for legal work: a workspace where a firm's documents are read, compared and drafted with the help of a language model. Documents are gathered into projects, questioned one at a time in the Assistant or many at once in a tabular review, and the answers point back to the passage they came from.

Darwin Axiom is a subscription-based AI system for the structured analysis, classification, summarisation and drafting of IP and legal documents. It extracts data, clauses, obligations, deadlines and legal concepts, and turns unstructured content into actionable information and finished documents. Manual review shrinks, consistency improves, and teams get far more done in the same time.

Knowledge sits in documents nobody has time to re-read: office actions, oppositions, licence agreements, coexistence agreements, assignments, opinions, contracts and correspondence. General-purpose AI fails professional work for three reasons:

• it cannot see your documents;
• it produces prose rather than a document;
• it rewrites rather than marks up, so nobody can see what changed.

Axiom addresses all three.

Axiom is for anyone whose work involves reading, comparing and producing IP and legal documents:

• IP agents and attorneys analysing office actions, oppositions and prior-rights evidence, and drafting responses, opinions and client reports.
• Law firms reviewing contracts, leases and agreements in volume, and drafting amendments as tracked changes.
• In-house lawyers and legal departments checking licences, coexistence and distribution agreements across the company's portfolio.
• Brand owners who want to understand the obligations, deadlines and risks hidden in the documents around their rights, without waiting for an external review.

No technical knowledge is assumed: questions are written in ordinary language and the answers come back the same way.

A project is a matter: a named place to keep the documents that belong together, with a client or matter reference and a practice area. Reviews and conversations can be started inside it, and a document drafted in one of them is filed there. Work does not have to sit in a project -- a document can also live on its own in Files -- but a project is what keeps one matter's material in one place.

Share a project with a colleague and they get everything in it, in one action: documents, folders, assistant conversations and review grids. The next person on the file inherits the thinking, not just the documents. A document can stay private even inside a shared project. When you share, Axiom stops and names any private documents before publishing them. Sharing never leaves your organisation: the check is enforced in the database, not on the screen. Folders can go to any depth, and projects can be filtered by practice area.

The Assistant is a conversation about the documents attached to it. It reads them, answers in ordinary language and cites the passages it relied on, so an answer can be checked rather than taken on trust. Documents can be attached as the conversation goes on, and each question is answered against what was attached when it was asked -- the transcript stays a truthful record of what each answer was about.

The Assistant works with a fixed, visible set of actions:

• it reads a document or searches inside it;
• it drafts a new Word, Excel or PowerPoint file;
• it amends an existing Word document as tracked changes;
• it copies a document so the original is untouched;
• it consults your organisation's own analysis templates;
• it asks you a question when it needs a decision.

Every action appears as a line in the transcript, so the work is visible, not implied.

Yes, in the Assistant and in a review's conversation alike. Ask for a document and Axiom writes it, files it where the conversation belongs and offers it for download. Ask for a change and it proposes one as a tracked change instead of overwriting the file: the proposal is read change by change, each accepted or rejected, and every version remains available afterwards.

Each sitting produces a version labelled with how it arose: uploaded, generated, amended by the assistant, accepted or rejected. A partner can always see the earlier state and who moved it, and the document can go to the other side with its history intact.

A grid. The rows are documents and the columns are questions, and Axiom fills every cell by reading that document for that question. It is the answer to "what does each of these fifty contracts say about termination", asked once instead of fifty times. Each cell holds a short answer, the reasoning behind it and a pointer to the passage it rests on, so a surprising result can be traced to its source.

The work happens in four steps:

• choose the documents, which become the rows;
• write the questions, which become the columns (Axiom can suggest the wording);
• Axiom fills the grid row by row, citing every answer;
• talk to the grid.

Ask which leases break before 2030, which licences are governed by Portuguese law, or which coexistence agreements cover class 25, and have the answer drafted as a memo, schedule or spreadsheet in the project. Interrupted runs resume, and an unreadable scan is marked while the other documents continue. Anything drafted from a private document stays private, automatically.

Every organisation has a house method, and it usually lives in one person's head or in a Word file nobody can find. In Knowledge you write it down as templates:

• an assistant template starts a conversation with your own instructions, for example "review this licence as we review licences: territory, term, royalties, termination, in that order";
• a tabular template starts a review with your own questions.

A saved starting point. A tabular template holds a set of columns; an assistant template holds the instructions a conversation should follow. Starting from one copies what it holds, so the thought that went into a good set of questions is spent once and reused on every matter after it. The copy is independent -- editing the template later leaves reviews already created from it untouched.

Axiom ships a starting set. You can hide any shipped template for your organisation, reversibly, or write your own from scratch. Templates carry a language, a practice area and optional jurisdictions, and can be shared read-only, or editable, with named colleagues.

The Axiom dashboard has three tabs:

• Operations: how much work was done and how it divides.
• Artificial intelligence: what the models cost, by supplier and by feature.
• Team: who is on the account.

Where Darwin, the suite's own supplier, meters usage, the figures are exact, because the tokens come out of your plan's allowance. Where you use your own Anthropic, OpenAI or Gemini key, Axiom shows its own count and says plainly that the provider's console is authoritative. Nothing on the dashboard is generated by AI.

A credit is a token actually consumed by the model, counted after each answer rather than estimated before it, so a short question costs less than a long one. The monthly allowance comes with the subscription plan and is shared by everyone in the account. When the balance reaches zero, the actions that call a model stop until the allowance renews or more is added; nothing already produced is lost, and everything already on screen stays readable.

• HTTPS only. The organisation and user are identified from the signed-in session, never from anything a request could alter.
• Documents are stored per organisation, and the reading mechanism cannot reach outside that area.
• The model receives only the text of the documents in the conversation and the question: never other matters, other users' private documents or other clients' data.
• Your own AI provider keys are encrypted before storage and never shown again in clear.
• Nothing is overwritten: amendments are additive and every version is kept.

It keeps the document to the person who put it there. Nobody else in the account can open it, download it or see it quoted, and it drops out of the rows of any review a colleague looks at, taking the answers extracted from it with it. The restriction follows the document rather than a copy of it: every version, every proposed change and any document drafted out of it inherits the same restriction.