KiwiStack

№ 01.8 / Office Suite — Intelligence

Intelligence

№ 01.8 · part of the suite

A dedicated AI workspace alongside the suite — search, ask, summarise, draft.


What it does

A dedicated AI workspace alongside the suite — separate from the apps but plugged into them. Ask in plain language, get answers across your suite with citations you can click through to. Each role can have its own assistant scoped to the data it should see. Powered by Onyx (the open-source AI workspace) and Mistral La Plateforme, both EU-hosted. Voice mode and multi-model side-by-side included.


Onyx chat surface with a generated answer and a sources sidebar listing the documents used to ground the answer

Onyx 3.x · Onyx · via docs.onyx.app

№ 01 · Featured

Cross-suite answers with citations

'Find the Vendor X contract' — the workspace returns an answer with citation chips below it. Click any source to open the original. No more guessing which app the document was in.

✓ Part of the Intelligence add-on

Onyx custom-agent configuration with system prompt, knowledge sources and actions

Onyx 3.x agents · Onyx · via docs.onyx.app

№ 02 · Featured

Custom Assistants per role

Sales gets a sales assistant scoped to the CRM and shared sales folder. Legal gets one scoped to contracts. Support gets one tied to the ticket archive. Each persona has its own data scope and tone — admins configure them once, the team uses them.

✓ Part of the Intelligence add-on

Onyx chat input with the file-context flow — adding files or URLs as context for the next question

Onyx 3.x file context · Onyx · via docs.onyx.app

№ 03 · Featured

Drop a document, ask anything

Upload a PDF, contract or Word doc into the workspace — get a one-page summary, ask for a more formal rewrite, translate it, ask any question. The original stays in your tenant; the workspace cites which page each answer came from.

✓ Part of the Intelligence add-on

Mistral Studio speech-to-text view of an audio file with timestamped segments, speaker-coloured avatars and a waveform timeline

Mistral Studio · Voxtral Transcribe 2 · via mistral.ai

№ 04 · Featured

Meeting transcription and search

After every videoconference, Voxtral — Mistral's audio model, EU-hosted — produces a high-fidelity transcript with speaker labels and word-level timestamps. The transcript lands in your workspace alongside mail, files and wiki; ask 'what did I commit to in yesterday's standup?' and get the answer with a citation back to the moment in the meeting.

✓ Part of the Intelligence add-on

AI-assistant sidebar in a videoconference UI listing action items extracted from the meeting — each entry names the owner that speaker committed to ('For Darnell to implement…', 'For Geeta to share…', 'For Porter to include…')

Brave Talk + Leo · Brave (same Jitsi + 8x8 backend) · via brave.com

№ 05 · Featured

Meeting minutes and action items

Once the transcript exists, Mistral derives the minutes: a short summary, an action-items checklist with the owner each speaker actually named, and a timestamped table of contents that links back to the moment in the recording. Skip to a topic, see what was decided, see what you committed to.

✓ Part of the Intelligence add-on

Jitsi Meet videoconference with the participant on screen, a yellow caption ribbon reading 'it is important that the video call has automatic subtitles', and the full call-control bar at the bottom

Jitsi Meet · jitsi.org

№ 06 · Featured

Live captions during the meeting

Real-time captions in the videoconference UI for accessibility, hard-of-hearing colleagues and mixed-language calls. Runs on your tenant cluster — audio never leaves.

✓ Part of the Intelligence add-on


Also included

The operational layer.

Web search alongside your data

Onyx Deep Research can pull from the open web during a chat — alongside your indexed sources, with the same citation discipline.

Voice Mode — chat hands-free

Speak the question, hear the answer; useful in the car, in the lab, while cooking dinner.

Multi-model side-by-side

Send the same prompt to two models in parallel — compare hosted Mistral vs. a local model when sovereignty matters.

Source citations on every answer

Every claim links back to the document or message it came from; one click to verify.

Per-team opt-out and per-customer rate limits

Admins choose who sees the workspace; LiteLLM caps per-customer and per-user usage.

DPA addendum specifically covering AI data flows

What's sent to Mistral, retention windows, your rights — written into the contract before you turn it on.

One thing to know

Notes intentionally has no AI. Notes is end-to-end encrypted — encryption keys live in the browser, the server can't read your notes, and indexing them in the workspace would break that property. If you need AI on a sensitive document, drop it into the workspace directly (it stays inside your tenant) or write it in Files where indexing is supported.


Available as

An add-on — +€10 / user / month.

Intelligence costs +€10 / user / month, opt-in per user or per team. Core, Mesh and Fleet base pricing is unchanged whether you take it or not. Turn it on at signup or later.