Your data lives elsewhere
Contracts and design documents end up in an outside provider’s storage. When you want them gone, you are not deleting them — you are asking someone else to.
A Levelupsoft Inc. software product
Your work conversations
never leave your servers
SSemTalk is a messenger your company runs itself. Messages and files are kept on your own servers, and your people use the messenger screens they already know to talk with colleagues and with AI agents your company builds.
- 4
- Desktop and mobile apps
- 10
- Languages supported
- AI
- Agents by text and voice (pilot)
Why self-hosted
What is convenience costing you
Public messengers are quick to adopt. In return, your message history, your attachments and a relationship map close to your org chart all pile up on someone else’s infrastructure.
Compliance is hard to prove
In a financial, healthcare or public-sector audit, asked "where is that conversation kept", you have nothing to offer but an outside vendor’s terms of service.
The rules change
Pricing, retention periods and feature limits are not yours to set. The day they change, the whole organization has to fall in line.
Product
SSemTalk ships as server and apps together
We do not hand over server software and leave the apps to you. Apps for four platforms are built to face your server and come with it.
Server
Install it in your data center or on an in-house server. One Docker set brings up the server, the database and the certificates.
Apps
macOS, Windows, Android and iPhone. Ship them under your own name and icon, and install them alongside a public messenger on the same device. An Apple Watch app is in pilot.
Bots and integration
Use the Bot API you already know. Existing Python and Node bots work once you change the server address, and can grow into AI agents that answer calls.
Features
Everything work needs, from day one
Messages and files
One-to-one and group chats, read receipts, replies, files up to 2GB, photos and documents.
Voice calls
End-to-end encrypted calls between two people, with a relay (TURN) for networks that block the direct path.
Bots and automation
Create bots with @BotFather and connect internal systems through the HTTP Bot API and webhooks.
Admin console
Users, bots and storage, how sign-up codes are delivered, and device sign-out, all on one screen.
Sign-up control
Whether registration is open, how verification codes travel (SMS, email, admin console), blocked numbers and attempt limits.
Ten languages
The apps and the documentation come in ten languages including Korean, with right-to-left layout for Arabic.
AI agents
Talk with AI agents Pilot
Your people write to an AI agent your company built from the desktop, call it from a phone to carry on, and keep going on an Apple Watch when they are out. All of it runs on SSemTalk installed on your company’s servers.
One agent, every device
Call the agent you have been writing to. The call record stays in the same chat on every device.
Straight from the wrist
Call an agent from the Apple Watch app, no iPhone needed. For now it is a pilot app that only makes calls.
Keep the conversation as text
When a call ends, the conversation can be kept as text in your own chat only. The person who made the call chooses whether to keep it.
Agents your company builds
Create a bot with @BotFather, turn on voice, then set its instructions, greeting, voice and language. It speaks nine languages.
Connected to internal systems
During a call, the AI calls your company’s tools (Bot API) and MCP servers to look up orders and check schedules.
Audio to your own AI server
In stream mode, your company’s AI server receives the call audio directly. It can also run on a model key under your company’s own contract.
Your company decides where the data goes
Unlike calls between people, calls with an agent are not end-to-end encrypted. The voice gateway decodes the audio and hands it to an AI model. With the default engine (Google Gemini on Vertex AI), the call audio goes to Google in the USA (us-central1). In return, your company installs the server itself, and can use its own model key, have its own AI server receive the audio in stream mode, and turn agent calls on and off with the operator switches. It is not “everything stays inside” — it is a design in which your company chooses where things go.
Voice agents and the Apple Watch app are in a pilot phase and not yet generally available. Messaging, the apps, bots and self-hosted installation are ready to use now.
Architecture
The layout at a glance
Apps connect to the server over standard MTProto, and the server records to PostgreSQL by event sourcing. Call audio between people travels between the apps, and using AI agents adds a voice gateway as an optional part.
Diagram explained
- Left to right: apps → edge → server → storage. The arrows that carry messages and files all end inside the company boundary.
- Solid lines are the path messages and files take. The server receives them, records them and pushes them to connected devices.
- The orange dashed line is voice. Call audio between people does not pass through the server, and goes by way of the relay only on networks that block the direct path.
- Internal systems attach as bots. Beyond uses such as alerts and approvals, an AI agent in a call reaches internal systems through Bot API tools or MCP servers.
- Storage accumulates events. What changed, when and how stays on the record instead of being overwritten.
- Dashed borders mark optional parts for AI agents (pilot). The voice gateway decodes the audio and hands it to the model, so agent calls are not end-to-end encrypted. With the default engine the audio goes outside the company (Google, USA); in stream mode it goes to the company AI server (the red dashed line leaves the company).
- The watch app signs in to the server over HTTPS and sends call audio straight to the voice gateway’s watch entry point.
Deployment
From install to everyday use, in four steps
- 1
Bring up the server
One Docker Compose set. The server, PostgreSQL and the certificates come up together.
- 2
Domain and certificates
Point your company domain at it and HTTPS certificates are issued and renewed automatically.
- 3
Roll out the apps
Download and install the apps carrying your own brand. Store distribution is possible too.
- 4
Sign-up and operations
Set how people join in the admin console, and see users, bots and storage in the same place.
Technology
Built on proven parts
MTProto 2.0
The same transport protocol the public messenger uses. There is no need to build new apps.
Java 21 · Netty
Requests are handled on virtual threads. A single server carries thousands of connections.
PostgreSQL · event sourcing
Events accumulate instead of state being overwritten. Audit and recovery come from the same record.
More than 5,000 tests
Real clients connecting and exchanging messages are checked automatically as well.
Thinking about deploying SSemTalk?
Tell us your scale and your compliance requirements and we will propose a setup that fits. We can help with a trial installation too.
Contact usLevelupsoft Inc.