Ask Axis — the assistant that does the work

Q: Does it really query my live business database?

Yes, read-only. It writes a single SELECT that runs on your live data through the same guarded engine as Analytics, with masked fields returned as NULL and results capped at 1,000 rows.

Ask Axis answers questions and does work across your records, functions and reports — every action checked against your permissions, never trusted blindly.

Rings of light pulsing out from a core into ordered blocks

Part of AxisIQ's single business model — The intelligence layer (Understand the business). How the layers fit.

Ask Axis is the AI assistant inside AxisIQ, and it is pointed at the same live data your team works in: the orders, stock, customers and invoices that would otherwise be spread across an ERP, a spreadsheet and someone's inbox, plus the reports you run, the functions your team has written and the documents in your knowledge base. It answers questions and it carries out work — and everything it touches is checked against the permissions of the person who asked. This page covers what it can reach, how those checks happen, and where it deliberately stops.

Plenty of business software now ships a chat window that can quote the help centre back at you. Ask Axis is the other kind: an assistant with tools rather than a manual, acting on your records under your own permissions.

You can ask it the questions you would otherwise build a report for — "how did last week compare with the week before?", "what has been sitting in packing since Tuesday?" — and you can ask it to act: "raise an invoice against order 1042", "run the reorder function for the low-stock list".

How a message becomes work

Each message goes to the assistant with a set of tools attached. It chooses which to use, runs them against your live workspace, reads what comes back and answers — sometimes after several rounds of that, all inside one reply.

It shows its working. The tools used in a turn appear above the answer, and any step with an input and an output — a query, a function run, something that failed — expands to show exactly what was sent and what came back. When it runs an analytics query, the table and chart appear in the conversation rather than as a claim about what the numbers say.

It runs as you, never as a superuser

This is the load-bearing decision, so it is worth stating precisely. Every tool call is evaluated against the permissions of the person who typed the message, at the moment of the call. There is no elevated integration account underneath doing the real work.

Three consequences follow:

If a field is masked for you, it is never handed to the model. It cannot repeat what it was never shown, so a well-phrased prompt is not a route around your permission rules.
If an action is outside your grants, the tool call is refused and the refusal goes back to the model as the result. It has to relay that to you; it cannot quietly report success it did not have.
Anything it writes lands in the audit trail attributed to you, with the before and after state, so "the assistant did it" is a traceable sentence rather than a shrug.

Risky actions ask first. Anything that moves money, sends a message outside the company, invites someone or deletes data stops with a plain "… — go ahead?" before it runs. Your yes approves exactly that action with exactly those details, once; ask for anything different and it asks again.

We have written separately about why that matters more than a well-behaved system prompt: an assistant needs permissions, not good intentions.

What it can reach

The assistant works through families of tools, and it is only ever offered the ones your workspace has. Four matter most:

Your records. It can list your object types, describe the fields on one, search, open a single record, and create, update or delete — under the same rules that govern the screens. Your custom types and your own field names are part of what it understands. See Operations.

Read-only analytics. It writes SQL and runs it on the same guarded engine behind Analytics: a single read-only SELECT over your own object types, fields you cannot read returned as NULL, results capped at 1,000 rows under a hard time budget. That turns "which customers ordered less this quarter than last" into a query with a defensible answer instead of a plausible-sounding paragraph. A query you liked can be saved as a report for the rest of the team.

Your knowledge base. It searches by meaning across the documents your team has uploaded and the record types your team has opted into indexing, so an answer can be grounded in your own material — a supplier agreement, an internal procedure, last year's policy note. Every result is filtered by what you personally are allowed to open, so grounding never becomes leaking. More in Knowledge Base.

Your functions. It can list, read, write and run the JavaScript functions your team keeps in the workspace. An invocation happens under your identity, exactly as if you had pressed run yourself — which is also why it cannot use a function as a way around a permission you do not have. See Automation.

Building and acting. It can also add record types and fields, draft flows and roles, raise payment links, send messages through your connected email, SMS or WhatsApp account, and generate PDFs from your templates — and, through MCP connectors your organisation adds, reach other systems over HTTP.

What it is genuinely good at

One-off questions that do not deserve a report. Ask, instead of building a view you will use once.
Multi-step busywork. Three actions across three screens is one sentence here, and you can read back which three it performed.
New joiners. Someone in their first week does not need to know where the dispatch screen lives; they can ask, and watch how it was done.
Explaining your own data. "Why is this invoice still open?" is answered from the records and the documents, with the source in view.

What it is not

It holds no standing memory of your business. Your conversations are kept as private threads — only you can see yours, and you can reopen or delete them — and within a thread it has the earlier messages, but everything it knew about your data in a turn came from a tool call it had just made. Indexing for the knowledge base is the one thing that happens outside a conversation, and that is opt-in per record type and switched on by whoever manages the knowledge base — not something the assistant decides for itself.

It is also not infallible. A model can misread an ambiguous question, or write a query that answers something adjacent to what you meant — which is why the tools it ran, the SQL it wrote and the records it changed stay on screen. Treat it as a fast colleague whose working you can check.

And it will refuse things. A denied action stays denied however the request is worded, and we would rather it said so than improvised.

The model underneath, and what it costs

The assistant is not tied to one vendor. It runs on third-party LLM providers, and which one answers is a configuration choice rather than something baked into the product — if a provider gets slower, dearer or harder to reach, changing it is a setting. With nothing configured, the assistant reports itself as switched off rather than degrading into something that half-works.

Usage is metered per message: tokens and cost are recorded for every turn and rolled up, so the workspace can see what the assistant costs as it happens rather than at the end of the month. Each person can be given a monthly limit — a default for everyone or one per person — with a notice at 80% and a hard stop at the limit, so one runaway conversation cannot spend everyone's allowance.

Where it fits

Ask Axis is at its most useful when there is enough in the workspace for it to reason over: Operations is what it reads and writes, Analytics is the engine behind its numbers, Knowledge Base holds its grounding material, Automation is the code it can call, and Permissions decides all of it.

Comparing AI assistants? See how AxisIQ compares with general AI chatbots and add-on AI agents, and read what an AI brain for business is.

FAQ

Can Ask Axis see data I am not allowed to see? No. Every tool call runs against your own permissions, and fields masked from you are never sent to the model — so it cannot disclose them even if asked directly.

Can it change my data, or only answer questions? It can change data: create and update records, run your functions, save reports, raise payment links and send messages. Each write is checked against your permissions first and recorded in the audit trail with your name on it, and money, messages, invites and deletions wait for your explicit go-ahead.

Does it really query my live business database? Yes, read-only. It writes a single SELECT that runs on your live data through the same guarded engine as Analytics, with masked fields returned as NULL and results capped at 1,000 rows.

Can it answer from our own documents? Yes, if you have put them in the knowledge base. It searches by meaning across uploaded documents and any record types your team has enabled for indexing, and drops any result you personally could not open.

Does Ask Axis remember previous conversations? Yes, as private threads: each conversation is stored and only you can see, reopen or delete it. Within a thread it has the earlier messages; facts about your business are fetched fresh by tool calls each turn, not remembered.