Capabilities AI-assisted

Models propose. Rules validate. People approve where judgement matters.

A language model reads the hard pairs a rule leaves open, drafts standardisation and enrichment, writes rules from a sentence and answers questions in plain words.

Proposematch · fix · enrich · classifyValidaterules · thresholds · securityReviewa steward, where it mattersApply and loggolden record · audit trailautonomy is earned per action type: a fix type that stewards approve 95% of the time can be allowed through, and can be taken backthe model learns from your stewards' decisions and your history; facts come from the ledger, and the model reads them thererow data stays in your estate; only structure and metadata cross to an external model
The loop every automated change goes through. Autonomy is earned per action type by approval history and can be taken back.
4
stages for every automated change
product
95%
approval rate before an action type can apply itself
product
0
rows of your data sent to an external model
product
1
sentence to draft a rule, tested on live rows before it is saved
product
Where models help

Six jobs, and what stays with people

JobWhat the model doesWhat a person does
Hard matchesreads two records with their context and says whether the two describe one company, and why, with a confidenceapproves merges of the kinds that matter; sets the thresholds
Standardisationknows that GmbH is a legal form, that St. can be Street or Saint, that a Japanese name leads with the family namereviews the exceptions the validator flags
Rules from sentencesturns 'open orders must stay under the credit limit' into a rule and shows the rows it would flagreads the rows, adjusts, saves
Enrichmentproposes a registry number, an industry code, a parent company from public and licensed sources, with the receiptapproves per source and attribute; the rule then applies within its range
Anomaliesflags a lead time that drifted from actual receipts, a value that changed too often, a supplier whose pattern changeddecides what to do; the flag becomes a rule if it recurs
Questionsanswers 'open orders for Acme in Q3 by plant' with governed SQL and citationsreads the answer with its rows; saves the query
The autonomy ladder

How an action type earns the right to apply itself

Level 0
Propose onlyEvery proposal waits for a person. The default for a new action type.
Level 1
Apply low-risk, reportActions with no transaction impact (a standardised spelling, a format fix) apply and appear in a daily digest.
Level 2
Apply within a range, review sampleActions with a defined range (an enrichment from an approved source) apply; a sample is reviewed weekly.
Level 3
Apply with an approval stampOnly after an evaluation set scores 95% or better, with escalation wired and a human approval stamp; revocable at any time.

The ladder is per action type, per domain. A merge stays at level 0 until the steward chooses otherwise; a spelling fix reaches level 1 in a week.

What crosses the boundary

Structure and metadata may cross; rows stay

A model needs the shape of the problem: field names, patterns, a rule text, an anonymised pair. Lima sends that. It sends no row data to an external model. Where a company wants a model to read rows, the model runs inside the estate, and the same boundary applies to it as to any user: the twelve security layers and the audit trail.

The trust centre states the boundary in one line: metadata can cross, row data stays.

Sent to an external modelfield names · patterns · rule text · anonymised examples
Kept in the estateevery row · every value · every key
Loggedevery call, its purpose and what it returned
Configurablewhich model, which provider, or none
Terms on this page

Action typeA kind of automated change: merge, spelling fix, enrichment from a named source, lead-time flag.
Evaluation setLabelled examples used to score an action type before it may apply itself.
Approval stampA named person's recorded approval that an action type may run at level 3.
Questions

Which model is used?

Any that the company approves: a hosted frontier model for structure and metadata, or a model running in the estate for row-level work. The choice is per action type.

How do we know why a model proposed something?

Every proposal carries its reasoning and its evidence, and the decision is logged with it.

Can we switch it off?

Yes, per action type and per domain; the rules and the queue keep working.

See your first domain built on your data

Bring two systems that hold the same customers, suppliers or materials. In one working day we connect them, run the match on live rows, review the queue with your steward and publish the first golden view with its API.