Bridge

A Bridge is the small service you run that sits between Teler and your AI provider. Teler opens a WebSocket to your Bridge for the call’s audio; the Bridge in turn opens a WebSocket to your AI provider (OpenAI Realtime, ElevenLabs, Gemini Live, etc.) and relays audio between them.

The Bridge owns:

• WebSocket lifecycle: connect, handshake, reconnect, close.
• Audio transport: base64-decode incoming, base64-encode outgoing.
• Sample-rate conversion if Teler and your provider use different rates.
• Interrupt and clear primitives for barge-in.

Your AI provider stays focused on speech-to-text, LLM, and text-to-speech.

Tip

You don’t need to build this from scratch. Every entry in Recipes ships a working Bridge as a Docker-compose repo for one provider. Clone, set env vars, and run.

Inbound call

A caller dials your Teler number, and Teler asks your Bridge what to do.

⤢

Step	What happens
1	Caller dials your Teler number.
2	Teler POSTs the Incoming Call URL on the Voice App.
3	Your Bridge returns a stream Call Flow with a ws_url.
4	Teler opens a WebSocket to that ws_url and starts streaming audio.
5	Your Bridge proxies audio to and from your AI provider.

Outbound call

Your code asks Teler to dial a number and connect it to your Bridge.

curl -X POST "https://api.frejun.ai/api/v1/voice/calls/initiate" \
  -H "Content-Type: application/json" \
  -H "X-API-Key: YOUR_API_KEY" \
  -d '{
    "from_number": "+91XXXXXXXXXX",
    "to_number":   "+91XXXXXXXXXX",
    "flow_url":    "https://your-bridge.example.com/flow",
    "status_callback_url": "https://your-bridge.example.com/webhook"
  }'

⤢

Step	What happens
1	Your code calls POST /api/v1/voice/calls/initiate with from, to, and flow_url.
2	Teler dials the callee.
3	When the callee answers, Teler POSTs your flow_url.
4	Your Bridge returns a stream flow with a ws_url.
5	Teler opens a WebSocket to that ws_url and starts streaming audio.
6	Your Bridge proxies audio to your AI provider.

Bridge responsibilities

Concern	Owned by Bridge	Owned by AI provider
WebSocket connect, ping, close	yes	no
Base64 decode/encode	yes	no
Sample-rate conversion	yes	no
Audio buffering and chunking	yes	no
Interrupt / clear on barge-in	yes (forwards to Teler)	yes (decides when)
Speech-to-text	no	yes
LLM inference	no	yes
Text-to-speech	no	yes

Next steps

WebSocket Messages The exact message types you'll send and receive.

Recipes Drop-in Bridges for OpenAI, ElevenLabs, Gemini, Vapi, and more.

SDKs StreamConnector: the SDK primitive for building a Bridge.