Skip to main content
GiniLytics IT Solutions logo—Matrix

In-product assistant

Gini Assistant: ask your workspace, inside your permissions

Some questions are faster asked than navigated to. Gini Assistant answers them inside the session you are already in — within the same permissions, the same company boundary and an audit record for anything it changes.

How it works, and what constrains it

Each item below is functionality in the product today.

No second login

The assistant runs behind the same session as the rest of the product. There is no separate credential and no separate access decision.

Your permissions, checked before any work

Each tool call verifies the caller's permission before touching the database, and the underlying service authorises again on its own.

Most questions never reach a model

A pattern-matching router sends direct lookups (your projects, your tasks, what is overdue) and aggregations (how many, what percentage, which project has most) straight to the database. One query, no model call.

A fixed set of approved capabilities

The assistant works through a defined set of actions agreed at startup — read your projects, search your tasks, add a comment, update an issue. Identity always comes from your session, never from the question.

Destructive actions are confirmed first

Anything destructive comes back as a signed confirmation covering the action, its arguments, the user, the company and an expiry — so a change happens because someone chose it, not because a question was phrased a certain way.

Writes are audited

Every non-read tool call is recorded, including the ones that failed.

Nothing sensitive in the logs

The assistant's metrics and audit records deliberately exclude question text, tool arguments, answers, and any user or tenant identifier.

Chat History for oversight

Questions asked of the assistant are available as their own permissioned report, so a company that wants visibility of how it is used simply grants it.

Inside the product, not bolted onto it

The assistant reaches the same services, with the same company scoping and the same authorisation, that every screen uses. That is what makes it safe to give to a whole team: an answer can only ever contain what the person asking was already entitled to see.

It is also built to be careful rather than quick. When a question calls for judgement, it takes the considered path rather than guessing from a pattern — because a confident wrong answer costs more than a slower right one.

  • Runs in the session the person is already signed in to
  • Permission checked before any work is done, and again by the service underneath
  • Every action recorded, and destructive ones confirmed first

The questions teams ask it

Most of the day-to-day ones are direct: what is assigned to me, which tasks are overdue, what is the status of a project, when is the next milestone. Those are answered straight from your own data in a single query.

So are the counting questions — how many open bugs, what percentage of this sprint is done, which project has the most reopened issues. And when someone needs judgement rather than a number, compare, summarise and prioritise are there too.

In-product assistant: common questions

Can the assistant see data I am not allowed to see?
No. It runs inside your own session, every action checks your permission before any work is done, and the service underneath authorises independently. Company scoping is enforced in the data layer, so an answer only ever contains records from your workspace that your role allows.
Can it change or delete things on its own?
No. Anything destructive comes back as a signed confirmation covering the action, its arguments, the user, the company and an expiry, and runs only once you confirm it. Every action it takes is written to the audit log.
Do questions about my tasks leave the workspace?
The questions teams ask most — my projects, my tasks, what is overdue, how many, what percentage — are answered directly from your own database in a single query. Open-ended questions such as compare and summarise are handled by the model, still inside the asker's permissions.
Can we see what people are asking?
Yes. Chat History is its own permissioned report, so a company that wants oversight of how the assistant is used grants it like any other report.

Related

Or go back to all features.

Try it against your own work

A workspace takes a couple of minutes to create. Put one real project in it and see whether this shape matches how your team already works.

Create your workspaceTalk to us

Already using Matrix? Log in, or compare the plans.