The assistant
The assistant answers a legal research question from the corpus and cites the material it used. The citation is the point — it exists so that the answer can be checked rather than trusted, and an answer that cannot be traced to a source should be treated as no answer at all.
Assistant
Is a duty of good faith implied into a long-term supply contract?
There is no general implied duty; the question turns on whether the contract is "relational". Ashcroft v Pentland Holdings (CA, 2019) sets out the criteria. Rennick Ltd v Kestrel Partners (HC, 2021) declined to imply the term on facts including a detailed express regime. Both are cited below with the relevant paragraphs.
Has Ashcroft been followed?
I can show you later decisions that cite it, but I cannot tell you whether it remains good law — there is no treatment analysis in this corpus. Check a citator.
It reads the corpus. It has no citator, so it cannot warn you that a decision has been overruled, reversed or heavily distinguished. An answer built on an authority that has since been overturned will be presented with exactly the same confidence as a correct one. This is the single most important limitation of the module.
The agents
| Agent | Does | Still needs a lawyer to |
|---|---|---|
| Review | Reads a document and flags terms against your playbook. | Decide what to do about each flag. |
| Extract | Pulls dates, parties, values and obligations out of a document. | Verify the extraction against the document. |
| Compare | Identifies differences between two documents. | Judge which differences matter. |
| Summarise | Produces a summary of a long document or judgment. | Read the original before relying on it. |
Each is a labour-saving device on a mechanical task. None of them exercises judgement, and the pattern in the right-hand column is not incidental — it is the whole basis on which the module is safe to use.
Assistant and agent usage appears in Admin → AI usage alongside every other module. Legal work tends to involve long documents, which is where cost accumulates.
What this does not do
The most significant gap. Verify elsewhere.
Output is material for a professional to evaluate.
The review agent flags; it does not produce a marked-up document.
Access is per module. There is no matter-level access control.
Questions
Is our material used to train models?
No.
Can the assistant see our clause library?
The drafter uses the library. The research assistant works from the corpus.
Can we restrict who uses it?
Through Module access, at the module level.