“Open critical defects with no test case linked”
A table of records
The count, every condition it applied, and the rows — sortable, exportable, and each one opens beside the answer without losing your place.
Oracle
Ask your project anything. It already knows.
Oracle is the AI inside Hawzu. Ask in plain English and get the records, a chart, an answer from your own documents — or the right form, open. And it shows you exactly what it understood.
Understood as
automated Automation status is Automated × the last 10 days Created on or after 2026-08-15 × never been executed Never executed × never been executed Never added to any run ×
18 test cases
TC-1042 Checkout · apply promo code at basket High
TC-1108 Login · SSO redirect returns to origin Medium
TC-1153 Cart · quantity limit rejects 100+ High
…and 15 more
Four different things come back
It works out what kind of answer your sentence wanted before it works out the answer. You never choose a mode.
“Automation over the last 30 days”
A chart
Asked about a shape rather than a list, it draws one. Change the chart type or the window afterwards and it redraws instantly — no model, no waiting.
“How do I set up a CI pipeline?”
An answer from your documents
From your own specifications in Canon, or from Hawzu's documentation, with the source it used cited underneath. When neither covers it, it says so.
“Let's create a test case”
The right form, open
That was not a question. Ten records can be started this way — it opens the real form beside you, blank and ready. It takes you there; you write it.
Every one of these is a real question it answers
Not a wishlist. Each one is a row in the test corpus this feature is measured against before it ships.
Every one verbatim from the frozen corpus this feature is measured against before it ships — not a wishlist.
It knows your project, not just its schema
Every other tool would need you to learn its field names. Oracle learns yours — read fresh from the project each time you ask, so a label you made this morning is askable this afternoon.
What Oracle is shown about this project
Folders
CheckoutLoginReportingPaymentsOnboarding and yours
Labels
smokeregressionflakyp1 and yours
People
PriyaSamAditiMarco and yours
Releases
August release2026.8.1Hotfix 3 and yours
Statuses
OpenIn ProgressBlockedClosed and yours
Custom fields
Test TypeComponentSprint and yours
· And never what is written inside a record — no test case title, no defect description, no comment.
It knows what it doesn't know
A plausible answer to an unanswerable question is the worst thing a tool like this can produce, because it comes back with rows and therefore looks like it worked.
Which tester is fastest?
Not something this data can answer.
Nothing in your project records it, so Oracle says so rather than ranking your team on a number it would have to invent.
not Priya · 4.2 cases/hr a leaderboard nobody measured
Show me the important ones
Which did you mean?
Vague in, question back — with the answers already prepared. Select one and it runs.
not 142 test cases one reading of three, picked for you
Defects assigned to Sarah
There is no member called Sarah.
Names are looked up against your project, never guessed at.
not 0 defects good news that isn't true
And all of this is already true
Not settings you turn on, and not promises. Each one is a consequence of how the thing is built.
-
It sees exactly what you see
Never one record more. A role that cannot open Defects cannot ask about them — and cannot reach them sideways either, through a condition that reads defects to answer about something else.
because your permissions filter the vocabulary before the question is planned, so what you cannot read is never offered
-
It never sees a record
It is shown what your project's fields are called and the values they hold — your folder names, your labels, your people. Never what is written inside a record: no test case title, no defect description, no comment.
because the model receives the catalogue and your question; the query runs afterwards, with no model anywhere in the path
-
It cannot make things up
A field that does not exist cannot be expressed, and a status that is not yours cannot be asked for. Names — people, folders, releases — are looked up, never guessed.
because the query's shape is generated from your live project, so an invalid field is impossible rather than rejected
-
Change your mind for free
Every condition is shown, resolved. Drop one, sort a column, ask for more rows, redraw the chart another way — all of it re-runs on the spot.
because a correction re-runs the plan you already have; only a new question is planned again
Thursday · 4:05 PM — cutting the release
“Are we sure nothing shipped untested?”
The old answer was a filter you built from memory, an export, and a spreadsheet you did not keep. The new one is the sentence you already said out loud — with every condition it applied printed above the result, so the person who asks you “are you sure?” gets to read the same thing you did.
Oracle, answered plainly
What is Hawzu Oracle?
Oracle is the AI inside Hawzu. You ask a question about your project in plain English and it answers with the matching records, a chart, an answer drawn from your own documents, or — if you were not asking a question at all — the right form, open and ready.
Do I need to learn a query syntax?
No. You type the question the way you would say it. If it is ambiguous you get a short question back with one-click answers, and if your data cannot answer it you are told so rather than handed a plausible-looking table.
Does my test data get sent to an AI model?
No record does. Oracle is shown what is queryable in your project — field names, the values they hold, your folder and label names — together with your question. It never receives a test case title, a defect description or a comment. Answers drawn from your Canon documents are the exception by design: there, the passages it cites are what it read.
Does it respect roles and project permissions?
Yes, and it can never be more permissive than the screen for the same records. Oracle reads its permissions from the same rules that guard each page, so if your role cannot open Defects it cannot ask about them — nor reach them through a condition that would have read them.
Can it invent a field or return made-up rows?
No. The query is built from your live project, so a field that does not exist cannot be expressed, and every value is checked against what your project actually uses. Names — people, folders, releases — are looked up rather than guessed, and an unknown one comes back as a refusal instead of an empty result.
Can Oracle build charts?
Yes, from a fixed vocabulary of 115 combinations plus four trends at daily granularity. Change the chart type or the time window afterwards and it redraws immediately without calling a model. Saved dashboards, curated insights and the project health score are Observatory.
What can I ask about?
Thirteen kinds of record — test cases, requirements, tasks, defects, releases, test runs, executions, and the settings records themselves such as labels, shared steps, parameters and custom fields — plus twenty ways they connect, like coverage, execution history, defect links and flakiness. Your own custom fields, folders, labels, releases and team members are part of the vocabulary too.
What are its limits?
Oracle returns records or counts them — there is no summing, averaging or grouping, which is Observatory's job. Conditions combine with “and” rather than “or”, and results are capped at five hundred rows, a hundred by default. Charts come from a fixed vocabulary rather than being generated freely. When a question falls outside any of these, it says so instead of returning something that looks close.
Is it a chatbot?
The full Oracle page is not: it answers one question at a time, and every answer stands on its own with its conditions printed above it. The floating assistant in the header does hold a short conversation, and hands anything that needs room — a chart, a form, a long table — to the page.
Asking is one of three ways to look
Ask Oracle The question you did not know you would have. You are here
Watch Observatory The ones you will ask every week and want on a dashboard. Take a look → Hand over Reports The ones you have to give someone, signed off. Take a look → Or read every AI surface in Hawzu — including the five things we deliberately did not make AI.