AI meeting minutes for government and public-sector bodies — Arabic and English, inside your jurisdiction
Arabic first, English kept
Most meetings in the region are not held in one language. Khulasa transcribes Arabic and English in the same meeting — even inside the same sentence — and keeps English product names, tools and terms in Latin script instead of rewriting them in Arabic letters. Choose Auto-detect, Arabic + English, English or Arabic when you request the recording; the summary, decisions and action items come back in the language that dominated the conversation, so an Arabic meeting yields Arabic minutes and a bilingual one keeps both.
We make no claim for specific dialects or for other languages, and the application itself is fully bilingual, with a right-to-left Arabic interface for the secretariat that reviews the minutes.
Deployed where the state requires
Public bodies rarely have the option of sending meeting audio to a cloud AI provider, and Khulasa does not ask them to. The On-Premise plan installs the whole system — the meeting bot, speech recognition, the language model that writes the minutes, and the storage — on your own servers, inside your own network, and it can run with no outbound connection at all.
Private Cloud is the same system as a dedicated, isolated deployment we run for your organisation alone, hosted where you need it; the location is agreed with you at setup, and UAE-resident or GCC-resident processing can be elected so meeting content is not transferred outside the chosen region.
On both, the speech and language models are self-hosted: no audio, transcript or summary is sent to an external AI API. The deployment options are described in full on the on-premise page.
- On-Premise — your servers, your network, no outbound connection required
- Private Cloud — dedicated and isolated, run by us, in the region agreed with you
- Both — self-hosted speech and language models, unlimited seats and recording hours, quoted per organisation
Minutes with decisions and owners
Every recording produces four things. The transcript is the full text of the meeting with speaker labels and timestamps, in Arabic and English exactly as spoken; click any line to jump to that moment in the audio. The summary is a clean, readable account of what was discussed, written in the language that dominated the meeting. Decisions and risks are pulled out as their own lists — what was agreed, and what was flagged — so nothing is lost in the transcript. Action items name who does what, with an owner and a due date where one was mentioned.
Together they are a draft of the minutes, not the minutes themselves: the output is not a verbatim record and is not certified, and it should be reviewed by the secretariat before it is circulated as an official minute. What it removes is the blank page and the hours of listening back.
- Transcript with speaker labels and timestamps, Arabic and English as spoken
- Summary in the language that dominated the meeting
- Decisions and risks as separate lists
- Action items with an owner and a due date where one was mentioned
Access, audit and retention
Organisation roles and workspace groups decide who can record and who can read; each recording is private to its owner by default and shared with a group or with named people. Public share links stay off until an admin enables them and expire by default. A tamper-evident audit log records who did what — including who accessed which recording — with CSV export for your own review, and both dedicated plans let staff sign in through your own identity provider over OIDC.
A retention policy erases recordings everywhere after the window you set, a legal hold exempts the ones that must be kept, emails and phone numbers can be redacted before analysis is stored, and anything can be deleted permanently at any time. Media is never handed out through a public storage URL — it is streamed through the API behind the user's session.
- Roles, groups and per-recording sharing
- Tamper-evident audit log with CSV export
- Retention policy, legal hold, redaction and permanent deletion
- Sign-in through your own identity provider (OIDC)
Open recording, every time
Khulasa never records silently. Participants see a guest named KhulasaBot (Recording) — or the name your organisation chose, always followed by “(Recording)” — muted and without video. As soon as it is admitted, and before it starts recording, it posts a notice in the meeting chat in English and then in Arabic, saying that the meeting is being recorded to produce a summary and linking to the published recording notice.
Anyone who would rather not be recorded can ask the organiser to remove the bot, and the recording ends the moment it is removed. We recommend that the chair also mentions the recording when the meeting opens, and that you check your own rules and local law before recording.
Microsoft Teams in a government tenant
Khulasa joins Microsoft Teams as an anonymous browser guest — it never signs in to a Microsoft account — so the only tenant setting that matters is whether anonymous users may join meetings, which is allowed by default. If your tenant admin has turned it off, the join button never appears and the recording fails; ask them to allow anonymous join for the meetings you record.
Teams normally sends guests to the lobby, where the organiser or a presenter admits KhulasaBot (Recording) within five minutes; some tenants label external meeting bots and ask for confirmation, and admitting it is enough.
No Microsoft 365 Copilot or Teams Premium licence is needed, nothing is installed in the tenant and no Microsoft Graph permission is granted to the bot. Channel meetings limited to organisation members, live events, town halls and webinars with registration are not supported.
What we do not claim
Khulasa holds no NCA, DGA or PDPL certification, and no data-residency certification. We describe the deployment posture — where the data lives, what runs on your hardware, what is encrypted, who can see what — and the assessment against your own framework is yours to make.
A Data Processing Addendum is available, which we sign as your processor, and our sub-processor list is published on the Legal page. Every accuracy figure we publish was measured on the self-hosted configuration; the cloud tiers have not been benchmarked, and we claim nothing for them either way. Dedicated deployments are priced per organisation, not per seat, and sized with you.
- No NCA, DGA or PDPL certification is claimed
- Data Processing Addendum available; sub-processors listed on the Legal page
- Accuracy figures describe the self-hosted configuration only; cloud tiers not benchmarked
- Priced per organisation
Data Processing AddendumSub-processor listCompare every plan
Questions from public-sector teams, answered
Can the whole system run inside a government network with no internet access?
The On-Premise plan installs the meeting bot, speech recognition, analysis and storage on your own hardware, and it can run with no outbound connection at all: nothing in the processing needs the internet. To record a Microsoft Teams, Google Meet or Zoom meeting, the bot joins that meeting from inside your deployment the way any participant does, and that meeting connection is all that leaves your network. No audio, transcript or summary is sent to an external AI provider.
Where is the data stored and who operates it?
On the On-Premise plan the data sits on your own hardware, inside your own network, and the deployment is operated as agreed in your contract, with dedicated onboarding and support from us. On the Private Cloud plan the deployment is dedicated and isolated to your organisation, run by us, and hosted where you need it — the location is agreed at setup, and UAE-resident or GCC-resident processing can be elected. On both, deleted media is erased immediately, and database records can persist in encrypted backups for at most fourteen days.
Are Arabic minutes produced automatically?
Yes. When the recording finishes, Khulasa produces a transcript, a summary, decisions, risks and action items in the language that dominated the meeting, so an Arabic meeting yields Arabic minutes, with English terms kept in Latin script. They appear in the dashboard and the summary is emailed to whoever requested the recording. Treat them as a draft: the output is not a verbatim record and is not certified, and the secretariat should review it before it is issued as an official minute.
Is there an audit trail of who accessed a recording?
Yes. A tamper-evident audit log records who did what across the workspace, and it can be exported as CSV for your own review. Access itself is governed by organisation roles, workspace groups and per-recording sharing, and media is streamed through the API behind the user's session rather than through a public storage URL.
Do you hold NCA, DGA or PDPL certifications?
No. Khulasa claims no NCA, DGA or PDPL certification. What we offer is the deployment posture — on-premise or private cloud, self-hosted models, encryption in transit and at rest, isolation per organisation, no model training unless you opt in — together with a Data Processing Addendum and a published sub-processor list, so that your own assessment can be made against your own framework.
Does the bot need a Microsoft or Google account in our tenant?
No. On Microsoft Teams it joins as an anonymous browser guest, and on Zoom as a browser guest, without signing in to any account in your tenant. On Google Meet it joins from its own Google account and asks to join like any outside participant, so the host admits it; meetings restricted to a single Google Workspace organisation refuse outside accounts and cannot be recorded.