AI CaseQuery
From signing in to answers your firm can rely on

Getting Started

Cloud edition · 8 October 2026

AI CaseQuery answers questions about your firm's database in plain English, in a web browser. Nothing is installed on anyone's PC. It does not know on day one what your firm means by its own words — which files count as active, what referred means, how you count a sale. Getting started is connecting it safely, teaching it that, and then proving that it learned.

The short versionYour administrator signs in at your firm's CaseQuery web address and works down Settings › Setup checklist: connect your database read-only through your firm's connector, add your people, enter your AI provider key, and hide anything sensitive. CaseQuery then studies how your database is built and lists the words it needs you to define. Your leads settle those and check standard questions. Then you test it against figures from reports you already trust. After that, anyone who spots a wrong number sends it to a lead with one click.
1. Sign in, connect and protect Connector, people, AI key, Schema rules Your administrator, with your IT Day one: the Setup checklist 2. Prepare the database Done by your firm's connector Automatic, at one press Produces the list of words to define 3. Teach it your language Settle terms, check baseline questions Your leads The first weeks 4. Prove it with ground truth Figures from reports you trust Administrator, with your reports Once the main terms are settled 5. Keep it honest Users flag, leads settle Everyone Ongoing; re-check every quarter Grey: administrator · Purple: automatic · Green: your firm's people

Where to go next

Administrators and your IT: Admin / DBA Guide. Leads: Lead Guide: Fine-tuning. Everyone who asks questions: User Guide, and Working with Casey if your firm offers the assistant. All of them are under User guides in the left-hand menu.

Day one: the Setup checklist

Your administrator signs in at your firm's CaseQuery web address with the sign-in NetDirector provided, opens Settings, and starts on Setup checklist. It ticks off what CaseQuery can see for itself and asks the administrator to confirm the rest. Work down it in order:

  1. Connect your database. On Settings › Database. CaseQuery never connects to your database itself: your firm's connector does, read-only, and it connects out to CaseQuery, so no firewall port needs opening. If your database is on your own network, the connector goes on a Windows server near it: use Download the connector where this screen offers it, or ask NetDirector for it. Unzip it, double-click Install, and fill in the form that opens: press Create a pairing code (it works once, for 30 minutes), paste it in, and enter your database and a login that can only read. If your database is in a cloud service, NetDirector hosts the connector for you; your database must let the connector's addresses through its firewall. Either way, the database login stays with the connector and CaseQuery never sees it.
  2. Add your people. On Settings › Security, under People at your firm. Each person signs in with their own email address and password, as a User (asks questions), a Lead (also settles definitions and reviews answers) or an Administrator (also manages Settings and people). A new person's password is shown once, for you to give them privately.
  3. Enter your firm's AI provider key. On Settings › AI provider: OpenAI, Anthropic, AWS Bedrock or Microsoft Foundry, on your firm's own account. It is kept encrypted and applies to everyone at your firm.
  4. Save your Schema rules. On Settings › Schema rules, hide any column nobody should see, then save, even if you hide nothing. Until you do, CaseQuery reads only the database's structure, never its data. Do this before preparing, so preparation never reads those columns.
  5. Prepare the database. Press Prepare Database under Settings › DB Tuning (stage 2 below).
  6. Finish the list. Review Safeguards, decide about Casey on Assistant, and add any Business Rules if you have them; most come later from your own people.

Then your people can sign in and ask questions on Chat. Some of these steps ask the administrator to enter their password again first.

The five stages

StageWhat happensWhat you do
1. Sign in, connect and protectYour database is reached only through your firm's connector, read-only. It can never change your data.Work down Setup checklist, as above. Save Schema rules before stage 2.
2. Prepare the databaseIt studies how your database is built — its tables, counts and coded values — and your firm's connector does the reading, on your network or where NetDirector hosts it. No row of your data is sent to the AI provider. It writes a note for each coded column and lists the words your firm needs to define.Press Prepare Database under Settings › DB Tuning. Optionally add a data dictionary first.
3. Teach it your languageYour leads decide what your words mean, and check a set of standard questions your firm already knows the answers to.On Fine-tuning: settle each item in Definitions to settle, then mark answers in Baseline questions as Right or Wrong. A wrong answer goes back on the list to settle. An administrator can also fix a question to a verified query that answers the same way every time, under Settings › DB Tuning › Question-SQL pairs.
4. Prove it with ground truthIt answers questions you already have trusted figures for, and shows which figures it reproduced exactly. The figures are never changed to match an answer — a disagreement is the finding.Enter a handful of figures from reports your firm trusts under Settings › DB Tuning › Ground truth, then check the answers.
5. Keep it honestEvery wrong number anyone reports reaches a lead with their name on it, and settling it improves every later answer that depends on it.Users press Not right? under an answer. Re-run baseline questions and ground truth every quarter, and after large rule changes.

Two things people mix up

What these steps usePreparing the database is done by your firm's connector with plain SQL and asks nothing of your AI provider. A data dictionary you add is the exception: the dictionary itself, never your data, is read through your AI provider. Checking baseline questions and ground truth asks each question of your AI provider exactly as if someone had typed it, so it is billed to your firm's provider account. You are told how many questions a check will ask before you confirm.