Recent Updates to the Private Server: MFA, Groups, Voice and More

Here’s something districts that rent their AI rarely get: an update list they can actually plan around. The private server platform had two significant releases in October,. Most of what shipped isn’t flashy. It’s the plumbing that makes a private AI server for schools safer to run and easier to share across a district.

Some of it is flashy. There’s a realtime voice mode that talks back. I’ll get to that, and to the setting I’d want IT to review before students touch it.

These features are now available in the platform; districts get them as their server is updated through support. Below is what changed, translated into who uses it and what it looks like on a Tuesday in a real school.

Key takeaways

• Administrators can now require multi-factor sign-in for every account on the private server and sign a user out of all devices with one click.
• Nested groups let a district set access and default models once at the district level and have schools and departments inherit them.
• Shared assistants now keep a version history, so a bad edit can be rolled back instead of rebuilt.
• Realtime voice is powerful but relies on a commercial voice provider, so it’s a decision about cost, privacy and student access, not just a switch.

Security updates IT directors will want first

Required multi-factor sign-in. An admin can now require an authenticator app for all users. Everyone sets it up at their next sign-in, scans a QR code, enters a six-digit code and gets ten single-use recovery codes. This matters. CISA’s K-12 cybersecurity report named multi-factor authentication as one of the key investments districts should make, and stressed that security leadership has to come from the top, not just IT (K-12 Dive).

Two things for IT to know before flipping it on. First, turning on the requirement signs out every device, including the admin’s, so schedule it for a quiet afternoon, not 7:45 a.m. on a Monday. Second, separate switches let single sign-on users skip the extra step, which is worth deciding on deliberately if staff already sign in through your identity provider with its own MFA.

Sign out all devices. A new button ends every open session for a user. Think of a laptop that goes missing. One caveat from the release notes: API keys aren’t sessions, so they keep working after a sign-out. If you suspect an account is compromised, remove its keys too.

Nested groups. This is the one I’m most excited about for district-wide rollouts. A group can now sit inside a parent group and inherit what’s shared with it, including permissions and a default model for new chats. So you can build District → High School → English Department once. A new English teacher added to the department group automatically sees the district’s approved assistants, the high school’s tools and the department’s writing-feedback assistant. When several groups apply, the most specific one sets the default model.

What teachers and leaders will notice

Version history and rollback for shared assistants. Every saved change to a workspace assistant now creates a version, with the author and an optional note. If someone edits the district’s board policy assistant and its answers go sideways, an admin can make an older version live again. The platform even refuses to restore a version whose knowledge base or tools are no longer available.

A real file browser for knowledge bases. Knowledge bases now open like folders, with a preview of the original document next to the “indexed text” the model actually searches. People with write access can edit that indexed text directly. A curriculum coordinator who notices the assistant misreading a scanned pacing guide can fix the text the AI relies on without re-uploading anything.

Shared chats others can continue. When you share a chat, you can now choose “Allow replies” so colleagues continue the same conversation, with names, profile pictures and a typing indicator. Each reply runs under the sender’s own account and permissions, and the owner’s hidden instructions stay hidden. Think of an instructional coach starting a unit-planning chat and three grade-level teachers adding to it during a PLC.

Word and PowerPoint files, created on the server. The built-in code tool can now produce Word documents and PowerPoint decks using libraries bundled with the platform, so it works without internet access. A principal can turn meeting notes into a draft staff handout without the content leaving the building.

Realtime voice: great for some uses, a decision for students

The headline feature is realtime voice. In this mode, a voice model handles the conversation and hands real questions to the chat’s selected model, including its knowledge base. For a world language teacher practicing conversation, or a staff member who would rather talk through a problem than type it, this is impressive.

Here’s where I want to be honest. Realtime mode uses a commercial voice model from OpenAI, or Azure OpenAI as of the mid-October update. That means voice audio in those calls goes to that provider under its API terms, and the district pays the provider directly for the tokens used. Tokens are the small chunks of text (roughly three-quarters of a word) that providers bill by the million. That cost is not included in the server purchase. Calls also need a specific “Allow Call” permission, which is exactly where IT should pause. I’d turn voice on for staff first, check the provider’s current published pricing, and decide on student access with your privacy lead at the table.

The platform’s analytics now show input and output tokens by model and user, which makes it much easier to see what API use actually costs. I wrote about why paying for use beats paying per seat in AI Cost for School Districts: Paying for Seats vs. Paying for Use.

One experimental item, apps that join group channels, is off by default. I’d leave it there.

What district leaders can do this month

1. Write down your MFA decision. Whatever AI tools you use, decide whether every staff account requires multi-factor sign-in, and who handles lockouts.
2. Sketch your group tree. On one page, draw District → School → Department. Note which models and assistants each level should see.
3. Inventory your shared assistants. List who owns each one and how you’d recover it if it broke. If the answer is “rebuild from memory,” that’s a risk.
4. Make voice a policy question. Before anyone turns on AI voice for students, involve your privacy lead and decide which provider, which grades and why. Our podcast episode Beyond Screen-Time Bans: Student Safety in the Age of AI is a good starting conversation.
5. Train the people who run it. LeadershipAIplus offers an AI Server: User Guide and a Server Admin Certification. Features like nested groups are only as good as the people configuring them.

Why a private AI server for schools makes updates yours

When a consumer AI product ships an update, you find out what changed after it changed. When you own the platform, updates come to you as options. Your IT team decides when MFA turns on, whether voice is available and to whom, and which assistants each school sees. That’s the difference between renting a tool and owning a platform.

I recommend running the server on-site, inside the district network, where prompts and logs stay on hardware you own. The one-time purchase includes pre-configured hardware, 12 months of remote support and a year of virtual professional development, with no per-user licensing (see the FAQ). Open-source models run locally with no token fees. Commercial models connect only when and for whom you choose.

And this isn’t a vendor chasing seats. It’s offered through LeadershipAIplus, a nonprofit, educator-led program run by former K-12 administrators: educational leaders supporting educational leaders, now working with districts from Kansas to Washington to Berlin. If you’re still mapping the bigger picture, my planning guide to private AI servers for K-12 walks through the questions to answer first.

FAQ

Do districts with a private AI server get new features automatically?

New features become available in the platform, and districts get them as their server is updated through support. Your IT team still decides which ones to turn on and for whom.

Can a private AI server require multi-factor authentication?

Yes. Administrators can now require an authenticator app for every account, provide recovery codes, and sign a user out of all devices.

Does AI voice on a private server keep audio on district hardware?

Not in realtime mode. Realtime voice uses a commercial provider (OpenAI or Azure OpenAI) under its API terms, billed per token. That’s why it should be a deliberate choice, especially for students.

The bottom line

Good AI infrastructure is mostly boring: sign-ins, groups, backups, version history. October’s updates made the boring parts better, and the district gets to decide on the exciting part.

So here’s my question: if a feature showed up in your teachers’ AI tools tomorrow, who in your district would decide whether students get it?

If you want to talk through what these updates would look like in your district, schedule a Systems Review.

Sources

• Anna Merod, K-12 Dive, “CISA K-12 cybersecurity report,” Jan. 26, 2023. k12dive.com
• PrivatEDU AI, homepage and FAQ, accessed Oct. 31, 2026. privatedu.ai
• LeadershipAIplus, homepage and courses, accessed Oct. 31, 2026. leadershipaiplus.org
• Platform release notes for the October 2026 updates (Oct. 10 and Oct. 14, 2026), reviewed by the author.