A meter, not a surprise​

A small counter in Axi's chat input shows your consumption for the month — 640 / 1000, for example — and updates as you go. It changes tone as you approach the limit, so nobody hits the wall without seeing it coming. When your credits are used up, the chat input pauses with a clear notice until the monthly reset; everything else in Axiome — dashboards, queries, alerts, exports — keeps working exactly as before.

Talking with Axi uses your credits, and so does refreshing an AI summary. Summaries generated automatically for scheduled reports don't consume anyone's credits, and if your plan has no credit limit, the meter simply doesn't appear.

For administrators: usage, now in credits​

The Usage page now speaks the same unit: AI credits used across the workspace, per day and by model, and — on the This month view — each member's consumption as a share of their monthly allowance. The allowance is set by the workspace plan and applies per member; runs on your own API key (BYOK) never consume credits.

📖 Chatting with Axi · AI summary · Usage and AI credits

One setting, workspace-wide​

A new AI page in the admin settings presents each provider's lineup, from its fastest tier to its flagship — Claude Haiku, Sonnet, Opus and Fable; GPT-5.6 Luna, Terra and Sol; Mistral Large and Medium — each with a relative cost indicator. Every Axi conversation, dashboard build and analysis follows the choice — across all members, with nothing to configure individually.

Switching is immediate: new messages run on the new model right away, and conversations in progress simply continue on it. The Usage page already breaks spend down by model and provider, so you can compare the two after a switch.

AI summaries and alert analysis stay on the model Axiome pins for them, keeping shared summaries consistent for every reader.

📖 AI model · Usage

One less tab to open​

Until now, "how do I share a dashboard?" sent you here, to the documentation, while "what was revenue last month?" went to Axi. That split never made much sense: both are questions you have while working, and only one of them had an answer where you were.

Axi has now read every page of this documentation. Ask it "how do I schedule a report to Slack?", "what's the difference between a measure and a metric?", or "can I filter one tab without touching the others?", and it answers from the relevant page — and links it, so you can read the detail when you want it.

In your language, and honest about its limits​

Axi answers from the documentation written in your own language: ask in French and it reads the French pages and links the French URLs.

And when the documentation doesn't cover something, it says so. Product questions used to be where an assistant was most tempted to invent a plausible-sounding feature; now the answer comes from a page you can go and check.

It works everywhere Axi does — the home page, a dashboard panel, or the Ask Axi to edit action on a component.

📖 Meet Axi · Analyzing your data with Axi

Your provider account, your billing​

The AI admin page gains an API keys section, one slot per provider. Add the key for the provider of your selected model and every AI conversation, dashboard build and analysis runs on it from the next message — remove it and runs fall back to Axiome's key just as seamlessly.

Keys are verified with the provider the moment you save them, so a mistyped or revoked key is caught immediately rather than failing a conversation later. They are stored encrypted, never displayed again — only the last 4 characters remain visible for reference — and can be replaced or removed at any time by any administrator.

📖 AI model

AI summary component​

A fifth component type, alongside charts, tables, cards and text. Drop an AI summary on a tab and Axi reads every other component on it, then writes what the data says — what moved, what looks unusual, how things compare.

Pick the angles you want with Sections (Overview, Evolutions, Anomalies, Comparison), then Generate. The result is saved on the dashboard, so everyone sees the same text — stamped with when it was generated and by whom — rather than each reader triggering their own run. Change the tab afterwards and the component flags itself Outdated until you refresh it — your readers never see commentary that shifted underneath them.

One summary per tab, and it can also be delivered to Slack on a schedule.

Edit mode and View mode​

Dashboards now open in one of two modes, switched from the top bar. View is a reading mode: data stays live, but nothing can be dragged, resized or reconfigured by accident. Edit is the full toolbox.

If a dashboard was shared with you in View access, it stays in View — the toggle belongs to people who can edit.

📖 AI summary · Edit mode and View mode

Context​

Two free-form documents, in Settings → Account → Context:

• My context — your scope, your habits, how you like results presented. Only your own agent reads it.
• Company context — workspace-wide business knowledge, edited by administrators. Every member's Axi reads it, so a few good lines here improve everyone's answers at once.

Memory​

Axi also saves durable facts as it works. Say "remember that our fiscal year starts in February" and it keeps that — with the date it was saved, and optionally scoped to one data source.

Review and prune your own memories in Settings → Account → Memory. Company memories, shared across the workspace, are managed by administrators.

Both are handed to Axi as reference knowledge, not as instructions: text in a context document can't make Axi do something it otherwise wouldn't.

📖 Context and memory · Company context and memory

Threshold alerts​

Put a bell on any KPI card or chart and get a Slack message when its value crosses a limit you set. Six conditions are available — above, below, equals, crosses above, crosses below, and changes by a percentage — plus a debounce ("after N consecutive breaches") for metrics that jump around.

An alert notifies you once when it trips, stays quiet while the condition holds, and re-arms silently when the value returns to normal. A metric sitting below its target for a week won't flood your channel.

Scheduled reports​

The same feature, without a condition: a report posts on schedule whatever the numbers say. Three shapes are available — a single KPI, every KPI card on a dashboard, or an AI-written summary of a tab.

Where to find it​

Three ways in: the Alerts panel in a dashboard's left sidebar, the bell on a card or chart, or simply asking Axi — "alert me when revenue drops below 10k".

Delivery goes to a Slack channel or your own DM. An administrator connects the Slack workspace once, in Settings → Admin → Integrations; after that, alerting is self-service for everyone.

📖 Alerts and reports · Slack integration