Self-hostedAir-gap readyFull-stack SDKsAI Agents

Self-hosted
real-time voice, video & AI

SRTC is a real-time audio/video engine you can deploy entirely on your own infrastructure — with SDKs for Web, mobile, desktop and server, SIP/H.323/GB28181 interop, and a Python SDK that plugs any LLM into a live call.

  • Runs fully on-premises or in your private cloud — media never leaves your network
  • SDKs for Web, Android, iOS/macOS, Windows, HarmonyOS and WeChat Mini Programs, plus Python, Go and C on the server
  • Bridges legacy SIP / H.323 room systems and GB28181 cameras
  • Talk to a live AI voice agent right here — nothing to install
SAI Agent· Live

Press the button and start talking

Live captions appear here

5-minute session · microphone required · headphones recommended

≈50
large enterprises & listed companies served
1000+
concurrent talk streams per 16-core server (tested)
<400ms
end-to-end latency
30%
video packet loss tolerated
50%
audio packet loss tolerated
4K
ultra-HD video
2000+
participants per meeting
10+
SDK platforms
SAI Agent

Real-time AI voice agents

The agent you can talk to above is built on SRTC: the browser joins with the Web SDK, and a Python service brings the model into the same channel. The whole pipeline can run in your own data center.

Two pluggable AI pipelines

User speech
ASR
LLM
TTS
Agent voice
or
User speech
Speech-to-speech model
Agent voice
# Server side: pip install "srtc[pipecat]"
from srtc.pipecat_transport import SRTCParams, SRTCTransport

transport = SRTCTransport(token, SRTCParams(
    audio_in_enabled=True, audio_out_enabled=True))

pipeline = Pipeline([
    transport.input(), stt, llm, tts,
    transport.output(),
])

Low latency, interruptible

Replies in about a second, with barge-in; the demo shows live response latency

Fully self-hostable

Run both the RTC stack and the AI services inside your network

Pluggable

Cascaded or speech-to-speech — mix and match ASR, LLM and TTS

Joins meetings

The agent is just another participant, and hears each speaker separately by uid

Phone calls in and out

Connect phone lines over SIP via the Room Connector for AI phone support and outbound calling

Works with

asrSelf-hosted ASRDeepgramWhisperFunASR / Paraformer
llmSelf-hosted LLMOpenAIQwenDeepSeek
ttsSelf-hosted TTSElevenLabsCosyVoice
realtimeOpenAI RealtimeGemini LiveQwen OmniDoubao Realtime
Explore SAI Agent

Why SRTC

Built for teams that must own their infrastructure: private deployment, legacy-device interop and AI, in one stack

Truly self-hosted

Deploy the entire stack — clients and servers — inside your own network: private, public or hybrid cloud, with an edge gateway for crossing network boundaries.

Air-gapped capableHybrid cloudEdge gateway

Runs where you do

Linux servers on x86 or ARM, installed with Docker or from an offline package for air-gapped sites. Data stays in the region and network you choose.

x86 / ARMDockerOffline installer

Keep your legacy devices

SIP / H.323 room systems, GB28181 and RTSP cameras, intercoms and desk phones join the same calls as Web and mobile users.

SIP / H.323GB28181RTSP

Interop with cloud meetings

Over SIP / H.323, the room connector bridges Zoom, Microsoft Teams and other cloud meeting services into the same call.

ZoomTeamsSIP / H.323

AI agents built in

The Python server SDK sends and receives channel audio directly — connect speech-to-speech models or any ASR / LLM / TTS combination.

Python SDKAny modelBarge-in

Built in-house, end to end

Capture, processing, encoding, transport, decoding and rendering are all our own, with H.264 / H.265 / AV1 / AVS3 support — smooth video at 30% and audio at 50% packet loss.

Full control180p–4K adaptiveAI noise suppression

Works across air-gapped networks and data diodes

Defense, critical infrastructure and OT / industrial networks often sit behind data diodes or cross-domain guards that pass only a few whitelisted protocols. SRTC puts a gateway on each side: the inner one talks to the platform, the outer one faces external users, and APIs, signaling and media are tunneled as a standard streaming protocol the guard already allows. Internal and external users share one meeting while the two networks never connect directly.

DefenseCritical infrastructureOT / ICSGovernment
External usersOuter gatewayData diode / guardInner gatewayPlatform inside
  • No generic TCP / UDP ports opened
  • Encrypted traffic passes through untouched
  • Three fixed tunnels, any number of users
  • One-way streaming out when required
How it works

Running on SRTC

Procuratorates, banks, public procurement, government and industry. Client names are withheld for confidentiality.

80+ customers Justice 20+ Government & public sector 20+ Education 20+ Enterprise 15+ Healthcare 6
Procuratorate

A provincial procuratorate

Video conferencing + VOD on a fully isolated private cloud

Private cloudConferencingVOD
Banking

A state-owned bank, provincial branch

Province-wide conferencing & training platform for all branches

ConferencingTrainingCustomization
Public procurement

A provincial public resource trading center

Remote bid opening & evaluation integrated with the trading platform

Remote evaluationIntegrationRecording
Security

A leading security company

Conferencing embedded in its procurement platform

SDK integrationCustomization
Energy

An energy group's e-commerce platform

Online negotiation meetings for centralized purchasing

ConferencingIntegration
Government

A municipal HR & social security bureau

Remote recruitment interview system

Remote interviewCustomization
Industrial IoT

An industrial technology company

GB28181 cameras connected via the device gateway

GB28181Device gateway
Unified comms

A communications equipment vendor

SIP / H.323 room systems interoperating with WebRTC

SIPH.323Legacy reuse

Architecture

A layered stack — adopt only the layers you need

Apps
STMLink Meeting |SLive streaming |SAI Agent
Extensions
Room connector |Cloud recording |AI add-ons (captions, minutes, virtual background, voice change)
Engine
SRTC Engine — client SDKs, server SDKs (Python / Go / C) and REST API
Protocols
WebRTC |SIP |H.323 |RTSP |GB28181 |RTMP
Deploy
Private cloud |Public cloud |Hybrid (edge gateway) |Tiered clusters with nearest-node access
SRTC platform architecture

SDKs for every platform

Web / H5
Android
iOS / macOS (Swift)
iOS (Objective-C)
HarmonyOS
Windows
C SDK (Linux / Windows)
Python (server-side AI)
Server API
WeChat Mini Program
Developer-first

A call in a few lines of code

Clean APIs, thorough docs and runnable samples. Client SDKs for Web, Android, iOS, macOS, Windows, HarmonyOS and WeChat Mini Programs; Python, Go and C SDKs plus a REST API on the server.

Complete docs, readable by AI coding tools (llms.txt / MCP)
Open-source demos for every platform
Direct access to our engineering team
import { SRTC, CameraPresets, MicPresets } from '@seastart/srtc-web-sdk'

// Initialize
const srtc = new SRTC()

// Join a channel
const channel = await srtc.join(token)

// Subscribe to the remote audio mix
const remoteAudio = await srtc.subscribeRemoteAudioMixTrack()
await remoteAudio.startPlay()

// Open the microphone
const micTrack = srtc.createLocalMicTrack(MicPresets.music)
await micTrack.startCapture()
await srtc.publishLocalTrack(micTrack)

// Open the camera and preview
const cameraTrack = srtc.createLocalCameraTrack(CameraPresets['720p'])
await cameraTrack.startCapture()
cameraTrack.addPlayView(document.querySelector('#local-video')!)
await srtc.publishLocalTrack(cameraTrack)

// Leave the channel
await srtc.leave()

All SDKs

PlatformInstallDocs
Web / H5npm i @seastart/srtc-web-sdkDocs →
Androidimplementation 'cn.seastart.rtc:rtc'Docs →
iOS / macOS (Swift)SPM: github.com/seastart/srtc-swift-sdkDocs →
iOS (Objective-C)pod 'RTCEngineKit'Docs →
HarmonyOSohpm: srtc-harmony.harDocs →
Windowsrtc-win-sdk (x86 / x64)Docs →
C SDK (Linux / Windows)librtc.so / .a / .dllDocs →
Python (server-side AI)pip install srtcDocs →
Server APIHTTPS + HMAC-SHA256Docs →
WeChat Mini Programnpm i @seastart/srtc-wx-sdkDocs →

FAQ

Common questions about SRTC

SRTC is a real-time audio/video engine by Hangzhou Freewind Technology that you can deploy entirely on your own infrastructure. It ships client SDKs for Web, Android, iOS/macOS, Windows, HarmonyOS and WeChat Mini Programs, plus Python, Go and C server SDKs and a REST API. Video adapts from 180p up to 4K across H.264, H.265, VP8, VP9, AV1 and AVS3. STMLink Meeting, SLive streaming and the SAI Agent conversational AI are built on top of it.

Ready to build?

Get a self-hosted trial, SDK access and hands-on help from our engineers