Four modes, one conversation, and it has read your files.
Most AI assistants know the sector in general and nothing about you. ELSA answers from your company profile, your policies and the documents you have uploaded, and switches between four ways of thinking depending on what you are asking for.
What this replaces today
- Asking a general-purpose chatbot a CQC question and getting an answer that is confident, plausible and about the wrong regulator.
- Searching your own document library by trying to remember what somebody called the file.
- Not asking at all, because explaining the context takes longer than doing the task.
- Advice that stops at advice, when what you needed was the task created and assigned.
How it works
Pick the mode, or just ask
Growth Advisor for the day to day. Compliance to quality-assure a care plan against Regulation 9, the Mental Capacity Act and nutrition standards. Strategy to score a tender and build a bid approach. Audit to test readiness against CQC's own published view and your records, with a plan sized to your inspection date.
It answers from your material
Your company profile, your policy library and your uploaded documents, retrieved by meaning rather than keyword, so the relevant passage surfaces even when you cannot remember the file name.
It can act, not only answer
Tasks created and categorised from the conversation, drafts produced and documents written, rather than a summary of what you should go and do.
What you get
It knows what you hold and what it says
Two different questions, answered two different ways: a roster of what is in your library by name, including drafts that are not yet searchable, and the retrieved content of the documents themselves. Asking what policies you have and asking what your medication policy says are not the same request.
Folder paths, not a flat list
ELSA reads the whole path a document lives at, Compliance then Audits then 2026, so it can tell one Audits folder from another and answer a question about a particular one.
Branch-aware
Somebody restricted to one branch gets answers from that branch. The file names and the quoted passages obey the same boundary, which sounds obvious and is exactly the kind of thing that is usually only half done.
Audit mode produces a plan, not a verdict
Audit starts from your CQC page, the ratings and the score CQC gave each quality statement, then checks your own records: reviews, sign-offs, incidents, missed visits, staff checks and policies. Each key question is graded by the evidence behind it, and the plan fits your inspection date, day by day if it is close. Where ELSA holds no record, it says so rather than guessing.
What stops it going wrong
Whether ELSA can read your library is a switch you own
Document reading is off by default for most roles and granted deliberately. It covers both halves, naming a file and quoting from one, because a file name on its own can be revealing enough.
Care records are never in scope
Documents attached to a person's care record are never indexed, so they cannot be surfaced in a chat answer no matter how the question is phrased.
It says when it found nothing
An empty search returns an explicit statement that no matching evidence was found, rather than a silent gap for the model to fill with something plausible.
Who it is for
Every plan. Strategy and Audit modes need Pro
For everyone in the service: a registered manager checking readiness, a bid writer scoring an opportunity, an administrator drafting a letter.
It gets better as your profile and library fill up, because that is what it answers from.