On-premise and private-cloud AI meeting notes for UAE and Saudi organisations
Two ways to keep meeting audio inside your perimeter
The three cloud plans process meetings with managed cloud models. The two dedicated plans replace those with self-hosted speech and language models, so your audio is never sent to a third-party AI provider. What separates them is who runs the hardware. Private Cloud is a dedicated, isolated deployment that we run for your organisation alone, hosted where you need it — the location is agreed with you when we set it up. On-Premise is the same system installed on your own hardware, inside your own network, and it can run with no outbound connection at all.
Everything else is identical: the same bot that joins Google Meet, Zoom and Microsoft Teams, the same Arabic and English transcription, the same summaries, decisions and action items, and the same workspace, roles and audit log your team would use on a cloud plan.
- Private Cloud — dedicated and isolated, run by us, hosted in the region agreed with you at setup
- On-Premise — installed on your servers, run inside your network, no outbound connection required
- Both — self-hosted speech and language models, unlimited seats and recording hours, quoted per organisation
Self-hosted speech and language models
The speech model is Whisper large-v3 from OpenAI, released under the MIT licence with open weights. We did not train it, and we say so first. What is ours is everything around it: the model runs at the published reference configuration its authors describe rather than at the library default, the language is decided once per recording instead of once per turn, and on the dedicated plans the whole pipeline — speech recognition, the self-hosted language model that writes the summary, decisions and action items, and the storage behind them — runs on hardware you control. No audio, transcript or summary is sent to an external AI API.
That is what the pricing page means by “self-hosted models” in the Private Cloud and On-Premise columns; the Free, Team and Business plans use managed cloud models instead.
The same bilingual notetaker, on your servers
A dedicated deployment does not change how a meeting is captured. Khulasa still joins as a named guest — KhulasaBot (Recording), or the name your organisation chose followed by “(Recording)” — from inside the deployment, records the room's audio, and posts a bilingual recording notice in the meeting chat. Nothing is installed in your Google Workspace, Zoom account or Microsoft 365 tenant, the bot needs no Microsoft or Zoom account, and nobody needs a Copilot or Zoom licence. Arabic and English are transcribed in the same meeting, even inside the same sentence, with English product names and terms kept in Latin script, and the summary, decisions and action items come back in the language that dominated the conversation.
The accuracy figures published on our home page were measured on exactly this self-hosted configuration, running on our own hardware: it is the configuration a dedicated deployment ships with, not an estimate for the hosted tiers.
Data residency, stated honestly
Private Cloud is hosted where you need it, and customers on Private Cloud and On-Premise may elect UAE-resident or GCC-resident processing, under which meeting content is not transferred outside the chosen region. On-Premise goes further: the data never leaves your hardware, because nothing in the system needs to leave it.
What we do not claim is a certification. Khulasa holds no PDPL, NCA or DGA certification, and we will not describe a deployment as compliant on your behalf — that assessment is yours to make, with our documentation in hand. What we can give you is the posture: encryption in transit and at rest, isolation per organisation, no model training unless you opt in, a Data Processing Addendum we sign as your processor, and a published sub-processor list, both on our Legal page.
When something is deleted, the media is erased immediately; copies of the associated database records can persist in encrypted backups for up to fourteen days, after which they are erased too.
- No PDPL, NCA or DGA certification is claimed — the deployment posture is what we document, and the assessment is yours
- Data Processing Addendum available; sub-processors listed on the Legal page
- Deleted media is erased immediately; encrypted backups hold database records for at most fourteen days
Data Processing AddendumSub-processor listMeeting data security
Governance you keep
The controls that decide who sees a meeting are the same on every plan, and on a dedicated deployment they are yours to run. Organisation roles and workspace groups decide who can record and who can read; each recording is private to its owner by default, shared with a group or with named people, and public share links stay off until an admin enables them and expire by default. A retention policy erases recordings everywhere after the window you set, a legal hold exempts the ones you must keep, emails and phone numbers can be redacted before analysis is stored, and a tamper-evident audit log with CSV export records who did what.
Both dedicated plans also let your people sign in through your own identity provider over OIDC, so accounts are created and revoked where you already manage them, and a REST API is available for your own integrations.
- Roles, groups and per-recording sharing, with access requests
- Retention policy, legal hold and redaction of emails and phone numbers
- Tamper-evident audit log with CSV export
- Your own identity provider (OIDC) — included on both dedicated plans
- REST API access
What is included and how it is priced
Both dedicated plans include unlimited seats and unlimited recording hours for every member, self-hosted speech and language models, and priority support with an SLA. Private Cloud adds the dedicated, isolated cloud environment we run for you; On-Premise is deployed on your own infrastructure and includes dedicated onboarding.
Neither is priced per seat: both are quoted per organisation and sized with you — how many people record, how much audio a month, what hardware you already have, and where it needs to live. Talk to sales and we will size it together. If a dedicated deployment is more than you need, every cloud plan still keeps recordings encrypted in transit and at rest, isolated per organisation, and out of any training set unless you opt in.
- Unlimited seats and recording hours
- Self-hosted speech and language models
- Priority support with an SLA — and dedicated onboarding on On-Premise
- Quoted per organisation — not per seat
Microsoft Teams, Zoom and Google Meet recordings that never leave
Microsoft Teams recordings and transcripts can stay on-premise: on the Private Cloud and On-Premise plans the bot joins the Teams meeting as an anonymous browser guest from inside your deployment, so the audio it records goes to your own storage and your own models, not to a third-party AI provider, and no Microsoft 365 Copilot licence is involved.
The same is true for Zoom, where Khulasa joins from the browser without a Zoom account or a paid Zoom licence, and for Google Meet, where it goes through the normal join flow and the host admits it. One dedicated deployment records all three platforms into one workspace, with one set of summaries.
Microsoft Teams AI notetakerZoom AI notetakerGoogle Meet AI notetaker
On-premise questions, answered
What is the difference between Private Cloud and On-Premise?
Both replace the managed cloud models with self-hosted speech and language models, so your audio is never sent to a third-party AI provider. Private Cloud is a dedicated, isolated deployment we run for you, hosted where you need it — the location is agreed with you when we set it up. On-Premise is the same system installed on your own hardware, able to run with no outbound connection at all. Both are quoted per organisation.
Does any audio leave our network on the On-Premise plan?
No. Recording a Google Meet, Zoom or Microsoft Teams meeting means the bot connects to that meeting the way any participant does; beyond that, the system needs no outbound connection at all. Speech recognition, analysis and storage run on your hardware, and no audio, transcript or summary is sent to an external AI provider.
Which speech and language models run on-premise?
The speech model is Whisper large-v3 from OpenAI — MIT licence, open weights — run at the published reference configuration its authors describe, with the language decided once per recording. A self-hosted language model writes the summary, decisions and action items. We did not train either model; what is ours is how they are run, and the fact that on the dedicated plans they run on your hardware.
Can we keep meeting data in the UAE or Saudi Arabia?
Yes. Private Cloud is hosted where you need it, and customers on Private Cloud and On-Premise may elect UAE-resident or GCC-resident processing, under which meeting content is not transferred outside the chosen region. On-Premise runs on your own hardware, so the data is wherever your servers are. We claim no data-residency certification: we document where the data lives, and the assessment against your rules is yours.
Is on-premise accuracy the same as the cloud tiers?
We cannot say, because we have not measured the cloud tiers. Every accuracy figure we publish describes the self-hosted, on-premise configuration — the one that runs on your own hardware — measured on real bilingual meeting audio. It is not a measurement of our hosted tiers, and we make no claim either way for them.
How is it priced?
Per organisation, not per seat. Both dedicated plans include unlimited seats and unlimited recording hours, self-hosted models and priority support with an SLA, and On-Premise adds dedicated onboarding; the quote depends on how many people record, how much audio a month, the hardware you already have, and where the deployment needs to live. Talk to sales and we will size it with you.
Do we still need Microsoft 365 Copilot or Zoom licences?
No. Khulasa does not use Microsoft Teams transcription, recording or Copilot: it joins as an anonymous browser guest, so nobody needs a Copilot or Microsoft Teams Premium licence and nothing is installed in your tenant. On Zoom it joins as a browser guest too, with neither Zoom cloud recording nor a paid Zoom licence, and it never signs in to a Zoom account.