3CX released V20 Update 10 on 11 November 2026, and it is the largest AI release the platform has had. For those of us selling and supporting 3CX across the UAE and the GCC, three things in it matter more than the rest: AI agents that behave predictably enough to put in front of real callers, a new way to interrogate your own call data, and transcription that can run on a machine in your own office rather than in somebody else’s cloud.
Here is what is actually in it, and what we think it changes for the projects we see.
Ask the phone system a question, in plain language
Update 10 adds support for Model Context Protocol, which lets you connect an AI assistant such as ChatGPT, Claude or Gemini to your 3CX system and ask it questions about your call data. Rather than exporting reports and comparing them by hand, a manager can ask why Friday afternoon had so many abandoned calls, which queues were affected and when the drop-offs started.
Two details are worth knowing before anyone gets nervous about it. Access runs through OAuth, and the connection respects the permissions the user already has in 3CX, so nobody sees call data they could not have opened in the console themselves. And this is a first release: 3CX has said more capability is coming, so treat it today as a reporting and troubleshooting aid rather than an administration tool.
AI agents that hold up on a real phone line
The AI Receptionist and AI Customer Service Agent have existed for a while. What Update 10 adds is reliability, which is the thing that decides whether a client keeps the feature switched on after the first month.
- Calls no longer disappear when the AI does. If the AI provider has an outage or an API key stops working, the call routes to a fallback destination you nominate, such as the support queue. This is the single most important change in the release, and it is the one we would demonstrate first.
- It finds the right person more often. Directory search handles larger organisations better, and copes when the caller gets a name slightly wrong.
- Long calls stay coherent. The agent condenses a lengthy conversation instead of losing the thread partway through.
- Instructions are followed more consistently, through improved underlying prompts.
The Customer Service Agent can now answer from your own documents, then hand off to a queue, ring group or extension when it cannot help or the caller asks for a person. A question about resetting a router gets answered; a billing dispute goes to accounts. Documents can be PDF, TXT, CSV or Markdown, and Markdown produces noticeably better answers than the others. Usefully, department managers can update their own documents without going through whoever owns the system.
There is also more provider choice: alongside OpenAI, agents can now run on xAI’s Grok. Costs between them were broadly comparable in 3CX’s own testing, so the sensible approach is to trial both against your real callers, in your own languages, and keep whichever handles them better.
Transcription that stays on your premises
This is the part we expect to matter most in this region. The new 3CX AI Server runs transcription, call summaries and sentiment analysis on your own hardware, either an Apple Silicon Mac or a Linux server. Call audio never leaves the building, and there is no per-minute charge. 3CX puts the throughput of a Mac mini at roughly 268,000 call minutes a month, which covers a great many contact centres comfortably.
For banks, healthcare providers, legal firms and government-adjacent organisations across the UAE and Saudi Arabia, that changes the conversation entirely. Where a cloud transcription service was previously a non-starter because recorded calls could not leave the organisation, an on-premise AI server removes the objection rather than arguing with it. The cost model helps too: a known hardware purchase instead of a per-minute line item that grows with call volume.
Setup is less work than it sounds. The AI Server establishes its own connection to 3CX, so there are no ports to forward, no firewall exceptions and no certificates to manage. Transcription covers the major languages, English, German, French, Spanish, Italian, Portuguese and Chinese among them, with others available.
The smaller things people will notice daily
- Choosing which number you call from. A user can pick which of their outbound numbers a call goes out on. For a business running two brands, or offices in two countries, this solves a recurring annoyance. An administrator enables it.
- A pop-out dialler that stays on top while the user works in a CRM or helpdesk, in Chrome and Edge.
- Call summaries worth reading. Summaries now carry a recap, notes and action items with owners, and administrators can write their own summary prompt to control length and focus. Speaker identification is more accurate and sentiment comes back faster.
- Conferences in fewer steps — press Conference, pick people or type numbers, and it starts, with one panel for muting, adding and removing participants.
- Softphone deployment via SSO. Users install the app, sign in with Google or Microsoft credentials, and configuration happens by itself. No QR codes, no provisioning files. This makes moving off the deprecated Electron app considerably less painful.
- Firmware management in one place. A new Phones page handles validated or manufacturer firmware, default versions per model, overrides for individual handsets and rollbacks, with scheduled updates and outdated handsets flagged on the dashboard. Partners can reuse custom provisioning templates across multiple customer systems, which is a genuine time saver for anyone managing a fleet.
Before you upgrade
Four things catch people out, and all four are avoidable:
- Zendesk and Salesforce integrations need reauthorising afterwards.
- If you use onboard AI transcription, move to the new 3CX AI Server before upgrading, not after.
- A customised call summary prompt will be overwritten by the new default. Copy it somewhere first and put it back.
- Update the client apps too, including Windows Softphone V5.8, or users will not see the calling improvements.
Which edition does this need?
The AI features are not in every edition. AI agents, transcription and the AI Server sit in the AI Edition, starting at 8 simultaneous calls, and a few of the enterprise pieces, including the onboard transcription server and inbuilt failover, need 16 simultaneous calls or more. Licensing is by simultaneous calls rather than per user, so the edition decides what the system can do and the call count decides how many people can be talking at once.
If you are not sure where a requirement lands, our full 3CX feature comparison lists all 77 published features against each of the five editions, and you can search it.
Where Zen fits
Zen Interactive Technologies is a 3CX Platinum Distributor for the UAE and the GCC. We supply the licences, the handsets and headsets to go with them, and the technical cover behind your deployment. If you are scoping an Update 10 project, particularly one where on-premise transcription is the reason it is viable at all, talk to us before you quote it. We will tell you the edition, the licence size and the hardware you need, and we hold stock in Dubai.