Skip to main content

Back-end automation (no audio)

Use the terminal example as the connection skeleton for a service. Check module availability first. Each token identifies one agent; a team integration needs independently authenticated clients.

When to use it​

Observe state for a dashboard, fetch history and contacts, or send SMS and presence commands. These are control operations and do not require audible media. Options.Media: nil selects SyntheticMediaFactory; it does not disable WebRTC.

What works without real audio​

The synthetic leg negotiates PCMA, sends A-law silence and counts incoming RTP. It is cgo-free and needs no sound device. Calls, mute, hold, DTMF, recording and conference controls still run on the server. cli.Stats() returns aggregate sent/received packet counts for active legs; counts do not prove audible quality.

For calls a human hears on an external phone, register, then set SetWebrtc(false) and SetAgentNumber(number). Confirm routing in state before dialing and restore it after each new registration. Registration requests WebRTC reachability regardless of capabilities; omitting "webrtc" does not make it a passive observer. A pure state reader should register only when its reference data are needed. Auto-answer is off unless explicitly enabled.

Adding real audio later​

Pass a factory func(callID string) (Media, error) as Options.Media. It creates one leg per call:

type Media interface {
    Answer(ctx context.Context, offer string, iceServers []*bcv1.IceServer) (string, error)
    Stats() (sent, received int64)
    Close() error
}

Apply the call's IceServers (Urls, Username, Credential) to your peer connection for NAT traversal. Return the SDP answer from Answer. The SDK closes legs when calls disappear, the stream ends, or Close runs. Your factory supplies microphone/speaker capture and playback; control and state code stay the same.

See also​

Client lifecycle · Intents · Errors & reconnects.