Trusted records. Real relationships. Ready for your AI.
Datum is master data management for corporate law firms. It finds every version of a client, matter and person across your systems, keeps one governed record of each, and maps how they connect: parent companies and groups, the parties on a matter, the people who work on it. Every field says where it came from.
Built for the agentic age. Your firm is about to have more AI assistants than lawyers reading its data, and an assistant answers with confidence from whatever it is given. Your reports, your conflicts engine and your AI agents should all read the same truth. Datum is where they get it.
master data management · corporate law firms · agents read here
- a record as a source system holds it
- the governed record, source attached
- a walled matter: reads as not found
The same client, four times, four spellings.
Your practice management system holds one version. The CRM holds another. A partner's spreadsheet has a third, and Companies House a fourth. Nothing in the firm decides which is right, so every report, every pitch and every AI answer is built on a guess.
Reporting nobody signs off
Client revenue is assembled by hand, because finance and the CRM disagree about which companies belong to the same group. The same reconciliation runs every quarter.
Experience nobody can find
A pitch team cannot show what the firm has done for a group. The matters sit under six spellings of the same parent company, and nothing joins them up.
Answers nobody should trust
An AI assistant asked about a client answers confidently from two records that were always the same one. It cannot tell, because nothing told it.
One governed record, and how everything connects.
Datum finds every version of a client, matter and person, works out which are the same, and builds one record under rules your firm writes. It holds the relationships between them, checks the quality of all of it, and serves the result to everything downstream, with its working shown.
Built field by field, under your rules
Each field of the record is chosen by a rule written in plain terms: prefer the register for the legal name, prefer the practice management system for the client-since date. A person can overrule any of them, and the overrule is recorded with a reason.
Every field says where it came from
One of your systems, an official register, a decision someone made, or Datum working it out from a match. Four kinds of claim, shown in four colours, never mixed. A guess from name similarity is never dressed up as a verified fact.
Every relationship, held as data
Which companies belong to which group, which parties are on which matter, and who at the firm works on it, at what rate, from when. Each link is dated and has a source, so "what have we done for this group?" and "who was on this matter in March?" are questions you ask, not projects you run.
Ask what you knew, and when you knew it
Datum keeps two clocks: what was true, and when the firm believed it. Ask what a record said on the day a matter opened and you get the answer the firm actually held then, not today's answer applied backwards.
Data quality is a queue, not a report.
Everything Datum finds wrong, uncertain or unfinished becomes a piece of work with an owner, a reason and a deadline. Nothing lands in a spreadsheet nobody opens. These are the surfaces your team will use every day.
Match review
Possible duplicates, with the evidence for and against each one added up line by line. Confirm, reject or defer with one key. Every decision can be undone.
Golden record
One record per client, company, matter and person, built under your rules, with the source and date of every field shown beside it.
Hierarchies and groups
Corporate structures held so every parent is one step away. Model a restructure before it happens. Any total respects the ethical wall.
Data quality rules
Written by your team in a simple expression language and checked as you type. What a rule catches arrives as work to do, ranked by seriousness and by record type.
Six queues, one shape
Match review, suspended clusters, escalations, data quality, hierarchy and drift. Built the same way, so someone new can clear two hundred items with a one-page guide.
Reference data and taxonomy
Code lists, mappings between systems and the legal industry's standard classification of work (SALI), managed as data. Anything an import cannot place goes to a person, never to a silent default.
Search that understands legal names
Ltd, Limited or LTD finds the same company. Permissions apply while the search runs, so nothing screened ever reaches the results.
Retention and legal hold
Schedules your records manager can read, a hold register that beats every schedule, and erasure that works even though history is kept.
How an agent reads Datum.
An AI agent is another reader of the firm's data. It gets the same record a person gets, through the same checks, with the same source attached. Below is what one exchange looks like, and how Trovato uses it.
Agents ask. Datum answers.
An agent asks through the Model Context Protocol (MCP), the open standard AI tools use to call other systems. Seven questions: look up a record, search, which client is this really, classify this piece of work, everything about a matter, the corporate group, and propose something new.
get_entitysearch_entitiesresolve_entityclassifyget_matter_contextget_corporate_grouppropose_contribution
The wall applies to machines.
An agent reads with its own identity and its own credentials, through exactly the same checks as a person. A matter behind an ethical wall looks to it like a matter that does not exist. There is no account anywhere that sees everything.
AI proposes. People decide.
The one thing an agent can write is a proposal. It lands in the review queue as a suggestion, and a person accepts or rejects it. Nothing an agent finds changes the record on its own.
POST /mcp Authorization: Bearer •••••• (trovato-agent · mdm:read) { "jsonrpc": "2.0", "id": 41, "method": "tools/call", "params": { "name": "get_matter_context", "arguments": { "matter_number": "24-0417", "as_at": "2026-03-14" } } }
{
"result": {
"matter_number": "24-0417",
"client": "Pemberton Holdings Limited",
"jurisdiction": "England and Wales",
"classification": "Corporate — Mergers & acquisitions",
"restricted": false,
"parties": [
{ "name": "Pemberton Holdings Limited", "role": "client" },
{ "name": "Cavendish Technologies Ltd", "role": "counterparty" }
],
"personnel": [
{ "name": "A. Wainwright", "role": "responsible partner",
"from": "2024-02-12", "to": null }
],
"provenance": {
"client": { "source": "companies_house", "selected_at": "2026-08-14T02:14:07Z" },
"jurisdiction": { "source": "elite3e", "selected_at": "2026-08-24T02:14:07Z" }
}
}
} A matter behind a wall and a matter that does not exist get the same reply, byte for byte. The transport cannot tell the agent the difference either.
{"result":{"isError":true,"content":[{"type":"text","text":"not found"}]}} Live in Trovato.
trovato.ai · legal operations platform · lawyers run matters with supervised AI agentsIntake, already filled
When a lawyer opens a matter in Trovato, the client, the parties and the governing jurisdiction come from Datum. Nobody retypes them, so nothing is mistyped. Retyping is where wrong data enters.
Agents ask back
Trovato's Matter Agents ask Datum for matter context as they work, with their own credentials. Every field comes back with its source, which is what Trovato's Citation Firewall needs before an answer can ship.
Findings return as proposals
What an agent learns from a document goes into Datum's review queue as a suggestion. A reviewer decides. The record never changes on an agent's say-so.
Datum pushes. Trovato never reaches into your network. The firm owns the schedule, can see what left, and can stop it leaving without asking anyone.
Two identities, screened differently. A directory identity publishes the list of matters, with walled ones marked restricted. A detail identity publishes matter content and is screened by every wall. A walled matter's name crosses; its content never does. Nothing that could be read as a clearance ever leaves.
Switch it off and the route does not exist.
Not disabled, absent. With the integration off there is no agent route in the software at all. The whole platform builds and passes every test that way, and proving it is one of the eight checks that block a release.
The wall is applied last, to everyone.
An ethical wall keeps the people on one side of a matter from seeing the other. Most systems enforce it with a setting somebody has to remember. Datum enforces it inside every request, and attacks it on every build.
Every request says who is asking
There is no way to read data without an identity, no total that skips the check, and no export route around the side. The unsafe question cannot be asked.
The screen always wins
Who may see something and who is screened off are worked out separately, and the block is applied last. No account gets past it: administrator, service account, elevated access, or agent.
Totals are attacked on every build
A screened matter must not be deducible from a count, a group total or an export. Datum runs those attacks on itself before every release, the way a penetration test would.
Set-up refuses a firm with no walls
A system with no walls loaded lets everything through, and nothing on screen would say so. The handover check refuses to sign the firm off in that state.
The record, as the steward sees it.
Pemberton Holdings Limited on Northgate Wells LLP, the demonstration firm. Every value names its source. The billing arrangement is withheld from this role and says so: a field you may not see and a field nobody filled in are different problems.
For the people who will run it.
Runs in your own data centre or as a managed single instance. Nothing here asks your team to replace what already works.
Connectors are settings files
The mapping, the checks and a rehearsal run that writes nothing ship together. A connector that has never been rehearsed cannot write. Working examples for Elite 3E and a generic CRM are included.
Six ways in, each says who is asking
Look up a record from another system's reference; the full record as at any date; search with permissions applied as it runs; what changed; bulk extracts that say which columns were withheld; the data model filtered to what you may see. A lookup answers in under 200 ms, nineteen times in twenty.
The database enforces the rules
Ninety-one tables in PostgreSQL 16. The database itself refuses to hold two conflicting current versions of the same thing, so application code cannot slip one past.
Joins onto what you run
Airflow, dlt, dbt, OpenSearch, Keycloak, OpenFGA and Marquez are joined onto, not rebuilt. Two AI connections are kept apart in the code, so a convenience for developers can never reach live client data.
The figures in this section are measured on our test set, not your data.
Datum is not a conflicts system and issues no clearance decisions.
It gives your conflicts system a clean register of parties to search: who they are, the groups they belong to, the other names they go by. Better input to the system you already run.
Nothing in Datum works out whether a matter can be taken, stores such a decision, or hands one back. No field could be mistaken for one. The line is drawn in the software as well as in the contract.
See it run on a firm that does not exist.
One hour on Northgate Wells LLP, an invented firm with invented clients. A real run of the software, three ethical walls, and two users who differ by exactly one thing. Bring technology, risk and leadership to the same hour. No customer data, no NDA.
Questions before booking? Use the contact form.