France is rolling out Visio, a homegrown secure videoconferencing platform, to all government employees by 2027.

[i47i@hachyderm.io]

@androcat @Yoshi

And here's the game-changer: modern FOSS AI chatbots can automate 60-80% of Tier 1 support queries right now.

FOSS LLM CHATBOT SOLUTIONS (Available Today):

1. AnythingLLM (MIT license, fully open)
• Runs ANY local LLM (no cloud/API needed)
• Built-in RAG for documentation
• Desktop + server deployment
• Works on CPU or GPU
• Cost: $0 + hardware

2. LobeChat (Open source)
• Multi-agent design
• Voice interaction (TTS/STT)
• Self-hosted, full privacy
• Multi-device sync

3. Open WebUI
• Lightweight, fast
• Ollama integration
• Perfect for quick helpdesk queries

BACKEND LLMs (Self-Hosted):
• Mistral 7B / MiMo-V2-Flash: Run on consumer hardware
• DeepSeek-V3.2: GPT-level reasoning, fully open (MIT)
• LLaMA 4: Fine-tune on your org's support tickets

THE MAGIC: These can be trained on:
• Linux documentation
• LibreOffice guides
• Your organization's procedures
• Historical support tickets

RESULT:
24/7 instant responses
Context-aware (knows user's history)
Handles common queries automatically
Escalates complex issues to humans
Zero per-query API costs
Complete privacy (no data sent to vendors)

REAL-WORLD EXAMPLE:
Manufacturing company case study showed chatbot deflected 70% of password reset/basic config tickets, freeing support staff for strategic work.

IMPLEMENTATION:
• Phase 1 (3 months): Deploy pilot with AnythingLLM + Mistral 7B
• Phase 2 (12 months): Scale to 60-80% Tier 1 automation
• ROI breakeven: 6-12 months

@androcat You're absolutely right—incremental "did you know" PSAs work. But imagine those delivered 24/7 by a chatbot that learns your patterns and proactively helps. That's available NOW with FOSS tools.

The technology exists. The economics work. Europe's proving it at government scale. The question is: do organizations have the will to invest in long-term self-sufficiency over short-term convenience?

[juliangonggrijp@ieji.de]

@i47i @barbarakathmann

[oceane@gotosocial.social]

@i47i I’ve just read a conversation between several Matrix users who agreed not to recommend it to non-technical users…

[phoenix@chaos.social]

@i47i Microsoft SQL server hits me every time.

Such a unnecessary piece of technology. Why would anyone want to pay a license for THAT?!?

[i47i@hachyderm.io]

@oceane

Yes. That's where local organisations we trust can setup ones like the one I'm using. #Matrix

[elettrona@poliversity.it]

@i47i @matrix There are alternatives for ALMOST everything. What about screen readers for blind users?
Windows has an open source screen reader called NVDA, which can be an alternative to the closed JAWS. But the Linux one, Orca, has still a lot of issues. And blind users have very few references about pros and cons.

[johnjburnsiii@kzoo.to]

@i47i

#Hmmm

They are using a Microsoft "diagramming/drawing" tool - as videoconferencing?

Guessing they just named it the same. But that will certainly result in confusion.

[i47i@hachyderm.io]

@JohnJBurnsIII

Microsoft Visio = diagramming/flowchart software (desktop application for creating technical drawings, org charts, floor plans, etc.)

France's "Visio" = videoconferencing platform (their sovereign alternative to Zoom/Teams)
This is definitely going to cause confusion, especially in international contexts. The name likely comes from:

[androcat@toot.cat]

@i47i @Yoshi

And users will hate using those AI chatbots exactly as much as they already hate the MS-based chatbots.

Value for money.

[elasticsoul@mastodon.social]

@i47i

Hey #Canada (and everyone non-US), the French police did this. What are our plans

"TIMELINE:
• 2005: Migrated from MS Office to OpenOffice
• 2008: Started Ubuntu desktop deployment (GendBuntu)
• 2014: Majority migration complete
• 2024: 97% of workstations running Linux (103,164 computers!)"

#Canada #cdnpoli #libreoffice #Linux #Ubuntu #ElbowsUp

[yoshi@toot.community]

@i47i I should have known there are AI chatbots for these sorts of "lookup" tasks. A system that can assist from simple newbie questions to more advanced users would be very helpful. The Linux world needs to better market these AI support systems. For example Linux Mint could "push" regular tips to users. I spent nearly 40 years in high tech from Radio Shack TRS-80 to huge VAX 11/780 systems but the current tech baffles me in many ways. Thanks to you and androcat for the tips.

@androcat

[yoshi@toot.community]

@androcat The "hate" could be a strong aversion to not knowing how and where their data are used. At least, that's why I look askance at anything MS or any other multi-billion dollar companies.

@i47i

[androcat@toot.cat]

@Yoshi @i47i

The trick really is to make it as transparent as possible.

Check if the question fits with a FAQ entry.
If yes: Link to FAQ and check if that resolves.
If any no: Go to ticket creation process (or live human chat if available)

People absolutely hate being given the "well-meaning runaround" that LLMs excel at. It's a waste of time and everyone knows it innately.

(And yes, the steps involved could just be a script, that's the general pattern for all things LLM - if they are good at it, it could be a script instead).

[yoshi@toot.community]

@androcat My take is biased by my level of understanding. I classify myself as "knows enough to be dangerous." The run around is about the same as one gets from clueless/powerless human support staff. I tried to get a refund from a human, it was a circle jerk of the first order.

@i47i

[androcat@toot.cat]

@Yoshi

Oh absolutely, but the human runaround happens because the human hasn't been empowered to perform the task (dark patterns).

The runaround with LLMs happen because they don't really understand anything and so aren't able to recognize what their own limits are.

If you ask them to do A, they will answer with a likely answer to someone asked to do A, "Yes, I will do that right now" - even if they do not actually have the API hookups to perform the action. A brain-dead yes-man.

@i47i

[yoshi@toot.community]

@androcat To the end-user, a runaround is a runaround, regardless of how it came to be. For me, with some level of understanding, I'd prefer a quick answer I can "fine tune" vs a slow answer that may still flawed.

@i47i

[i47i@hachyderm.io]

@Yoshi @androcat

I hate customer service chatbots as much as everyone else does. We're stuck with them, and I've had to learn how to work with them since there's simply no other option anymore. Many firms have moved to Messenger, WhatsApp, even Telegram for support.

But @androcat, your point about transparency is exactly right, and I just experienced it done well.

My bank recently changed their system to do precisely what you described: the chatbot does a quick FAQ check (and the LLM is actually quite effective at this), and if that doesn't resolve it, immediate escalation to a callback system - no waiting on hold, no runaround. A real human calls you back within minutes.

It respects your time instead of trapping you in conversation loops. The chatbot handles what it's actually good at (matching questions to FAQs), then gets out of the way for human support.

So it CAN work when designed with user needs in mind rather than just cost-cutting. The technology isn't the problem - it's whether companies have the will to implement it properly.

[mihamarkic@mastodon.social]

@i47i Slovenia? You sure? We don't even distinguish Word from word processor app.

[androcat@toot.cat]

@i47i @Yoshi

Yeah, everything Pattern Recognition flavored is pretty good.
Everything actually generative (let alone Agentive) is utter horsecrap.

[tomstoneham@dair-community.social]

@i47i
"The Tchap code is publicly available and is based on the Matrix protocol. The development of the application thus benefits from advances in the Matrix community, in terms of functional improvement and security."

But how do they *give back* to the @matrix #Matrix Foundation and Community?