Join a meeting
The live Atlas Video app — sign-in, rooms, calls.
- vc.click
Speech recognition, the language model and vector memory all run on our server in Moscow. Nothing goes to OpenAI or any other external service. For government customers and anyone bound by 152-FZ, this is what no cloud competitor offers.
Our own engine recognizes speech, a local language model answers, and our own vector memory stores the minutes — all on a server in Russia. No OpenAI and no foreign clouds: the content of meetings never leaves the perimeter. It is a property verifiable at acceptance — exactly what 152-FZ requires.
Our own engine, live subtitles right in the meeting, with nothing sent outside.
A local LLM parses the transcript and writes the minutes within the Russian perimeter.
Semantic search across past meetings; each organization sees only its own.
Speech is recognized by our own engine during the meeting, with no delay for sending it anywhere outside.
Each participant picks a language and sees colleagues’ remarks in it. A meeting with a foreign contractor no longer needs an interpreter.
When it ends, the model parses the full transcript and posts structured minutes to the chat: topics, decisions taken, action items with owners. No one keeps them by hand.
The minutes build up into a memory you can query in your own words — “when did we decide on moving the warehouse.” No exact wording needed. Each organization sees only its own.
An administrator uses a rule to describe what to do after a meeting: action items from the minutes turn into tasks in the ERP, go to a chat or to an external system. A builder, not a rigid integration.
Load-tested: 800 participants at once without a single connection error.
Our own ASR, our own synthesis, a local LLM and vector memory — from these building blocks.
“Atlas, record this decision” — the remark goes into the minutes. “What did we decide last time” — an answer from the meeting memory.
The assistant opens the meeting, announces the items and keeps to the schedule — it says when time is up.
Join midway and get a short recap instead of “tell me what I missed.”
“Why was the second option rejected,” “who owns the deadlines.” The model answers from the transcript, without making things up.
Not just the minutes text — jump to the exact minute of the recording by a phrase.
The video face and speech synthesis are ready; roles — webinar host, interviewer, on-duty consultant, auto-secretary on the phone.
At TrueConf, Vinteo and IVA the AI features are either absent or run through external clouds.
| Feature | TrueConf | Vinteo | IVA | Atlas Video |
|---|---|---|---|---|
| Live subtitles (RU) | limited | no | limited | yes, our own engine |
| Auto-translated subtitles | no | no | no | yes |
| AI meeting minutes | no / external | no | external cloud | yes, in-perimeter |
| Semantic meeting memory | no | no | no | yes |
| Outcome scenarios → ERP/chat | no | no | no | builder |
| Where the AI runs | — | — | external clouds | Russian jurisdiction, 152-FZ |
| Load | — | — | — | 800 participants |
The live Atlas Video app — sign-in, rooms, calls.
Atlas Video · a single AI core in Russia.
We’ll deploy it and show subtitles, minutes and memory in a real meeting.