AgentStream

Stream real-time audio between live phone calls and your bot server over WebSocket. Three ways to connect — pick the one that fits your setup.

How it works

Caller → Exophone → Exotel → wss://your-bot-server

When a call is answered, Exotel opens a WebSocket to your endpoint and streams raw PCM audio every ~100 ms. Send audio back on the same socket to speak to the caller.

Connect Voice AI

Dial a number and connect the answered call directly to your bot. No flow setup needed.

POST /v1/accounts/{account_sid}/calls/connect

Region	Base URL
Mumbai	`https://api.in.exotel.com`
Singapore	`https://api.exotel.com`

Auth: HTTP Basic — API key as username, API token as password.

Parameters

Parameter	Required	Description
`from`	Yes	Number to dial — E.164 format (`+919876543210`)
`callerid`	Yes	Your Exophone (shown as caller ID)
`streamurl`	Yes	Bot WebSocket URL — `wss://` or `ws://`, max 600 chars
`streamtype`	Yes	`bidirectional`
`record`	No	`true` to record the call
`recordingchannels`	No	`single` (merged) or `dual` (separate tracks)
`timelimit`	No	Max duration in seconds (max `14400`)
`customfield`	No	Metadata string, max 128 chars
`statuscallback`	No	Webhook URL for call status events
`statuscallbackevents[]`	No	`answered` · `terminal` · `ringing`
`streamname`	No	Label for the stream, max 32 chars

Request

curl -X POST \
  'https://<api_key>:<api_token>@api.in.exotel.com/v1/accounts/<account_sid>/calls/connect' \
  -F 'from=+919876543210' \
  -F 'callerid=08047491899' \
  -F 'streamurl=wss://bot.example.com/media' \
  -F 'streamtype=bidirectional' \
  -F 'statuscallback=https://your-server.com/callback' \
  -F 'statuscallbackevents[]=terminal'

Response

{
  "call": {
    "sid": "a1b2c3d4e5f6",
    "status": "in-progress",
    "from": "+919876543210",
    "phonenumbersid": "08047491899",
    "direction": "outbound-api",
    "datecreated": "2026-05-14 10:00:00",
    "recordingurl": null
  }
}

Status values: queued · in-progress · completed · failed · busy · no-answer

Connect Voice API with Flow

Dial a number and run the call through an Exotel flow — IVR menus, greetings, DTMF collection — before reaching your bot.

Same endpoint:

POST /v1/accounts/{account_sid}/calls/connect

Parameters

Parameter	Required	Description
`from`	Yes	Number to dial — E.164 format
`callerid`	Yes	Your Exophone
`url`	Yes	Flow URL: `https://my.exotel.com/{account_sid}/exoml/start_voice/{flow_id}`
`calltype`	No	`trans` for transactional calls
`timelimit`	No	Max duration in seconds (max `14400`)
`timeout`	No	Ring timeout in seconds
`statuscallback`	No	Webhook URL for status events
`statuscallbackevents`	No	`terminal` · `answered` · `both`
`customfield`	No	Passed into the flow via the Passthru applet

note

Don't pass streamurl or streamtype here — the WebSocket URL is configured inside the flow's Voicebot/Stream applet.

Request

curl -X POST \
  'https://<api_key>:<api_token>@api.in.exotel.com/v1/accounts/<account_sid>/calls/connect' \
  -d 'from=+919876543210' \
  -d 'callerid=08047491899' \
  -d 'url=https://my.exotel.com/<account_sid>/exoml/start_voice/<flow_id>' \
  -d 'statuscallback=https://your-server.com/callback' \
  -d 'statuscallbackevents=terminal'

Response

Same shape as Connect Voice AI response.

Common flow patterns

Pattern	Applets in flow
Greeting → bot	Play → Voicebot
IVR menu → bot	Gather (DTMF) → Voicebot
Bot → human agent	Voicebot → Transfer
Listen-only (transcription)	Stream → Passthru

Programmable Voice APIs (ExoML)

Full runtime control via Legs API — per-call stream URLs, dynamic routing, greeting + stream in parallel.

Step 1 — Dial the customer

curl -X POST \
  'https://<api_key>:<api_token>@api.in.exotel.com/v2/accounts/<account_sid>/legs' \
  -H 'Content-Type: application/json' \
  -d '{
    "contact_uri": "+919876543210",
    "exophone": "08047491899",
    "leg_event_endpoint": "grpc://your-event-server.example.com",
    "timeout": 30
  }'

Exotel emits: leg_connecting → leg_ringing → leg_answered

Step 2 — Start stream on `leg_answered`

curl -X POST \
  'https://<api_key>:<api_token>@api.in.exotel.com/v2/accounts/<account_sid>/legs/<leg_sid>/actions/start_stream' \
  -H 'Content-Type: application/json' \
  -d '{
    "direction": "bidirectional",
    "url": "wss://bot.example.com/stream",
    "content_type": "audio/x-mulaw;rate=8000"
  }'

Optional — Play greeting while stream initialises

Fire both requests simultaneously on leg_answered to eliminate dead air:

# Start say in parallel with start_stream
curl -X POST \
  '.../legs/<leg_sid>/actions/start_say' \
  -H 'Content-Type: application/json' \
  -d '{ "text": "Please hold.", "loop": 0 }'

Stop it as soon as stream_started fires:

curl -X POST '.../legs/<leg_sid>/actions/stop_say'

Scenario	Use greeting?
Inbound IVR / tech support	No
Outbound sales / collections	Yes

Applet configuration

Stream Applet — Unidirectional (Exotel → your server)

Audio flows one way only — from the caller to your server. Your server cannot send audio back. Use for transcription, agent assist, call monitoring.

#	Parameter	Required	Description
1	Action	Yes	`Start` to begin streaming · `Stop` to end it
2	URL	Yes	`wss://your-server.com/stream` — or `https://` endpoint returning `{"url":"wss://..."}`
3	Next Applet	No	Call flow continues to next applet immediately after stream creation

note

Unidirectional streams fork audio immediately. If used with a Connect applet that rings multiple agents, audio from all ringing legs is sent — filter on your end.

VoiceBot Applet — Bidirectional (Exotel ↔ your server)

Audio flows both ways — your server receives caller audio and can send audio back in real time. Use for conversational AI, IVR replacement, outbound bots.

#	Parameter	Required	Description
1	URL	Yes	`wss://bot.example.com/stream` — or `https://` endpoint returning `{"url":"wss://..."}` for dynamic routing
2	Authentication	No	IP whitelist or Basic auth header. Email hello@exotel.com for Exotel's IP ranges.
3	Sample Rate	No	`8000` (default) · `16000` (recommended) · `24000`. Append `?sample-rate=16000` to URL.
4	Custom Parameters	No	Up to 3 key-value pairs in URL, max 256 chars total. Arrive in the `start` event.
5	Record	No	Generates recording URL in the Passthru applet
6	Next Applet	No	Stream closes automatically — no Stop applet needed

WebSocket protocol

Event support by applet type

Event	Stream Applet (Unidirectional)	VoiceBot Applet (Bidirectional)
`connected`	✓	✓
`start`	✓	✓
`media` (receive)	✓	✓
`dtmf` (receive)	✗	✓
`mark` (receive)	✗	✓
`stop`	✓	✓
`media` (send)	✗	✓
`mark` (send)	✗	✓
`clear` (send)	✗	✓

Events — Exotel → your server

`connected`

{ "event": "connected" }

`start`

{
  "event": "start",
  "sequence_number": "1",
  "stream_sid": "MZxxxxxxxxxxxxxxxxxxxxxxxx",
  "start": {
    "stream_sid": "MZxxxxxxxxxxxxxxxxxxxxxxxx",
    "call_sid": "CAxxxxxxxxxxxxxxxxxxxxxxxx",
    "account_sid": "ACxxxxxxxxxxxxxxxxxxxxxxxx",
    "from": "+919876543210",
    "to": "+918047491899",
    "custom_parameters": { "key1": "value1" },
    "media_format": {
      "encoding": "audio/x-raw",
      "sample_rate": "8000",
      "bit_rate": "16"
    }
  }
}

`media`

{
  "event": "media",
  "sequence_number": "3",
  "stream_sid": "MZxxxxxxxxxxxxxxxxxxxxxxxx",
  "media": {
    "chunk": "2",
    "timestamp": "200",
    "payload": "<base64-encoded PCM>"
  }
}

`dtmf` (VoiceBot Applet only)

{
  "event": "dtmf",
  "sequence_number": "7",
  "stream_sid": "MZxxxxxxxxxxxxxxxxxxxxxxxx",
  "dtmf": { "digit": "5", "duration": "100" }
}

`mark` (VoiceBot Applet only)

Sent when audio you previously sent has finished playing.

{
  "event": "mark",
  "sequence_number": "15",
  "stream_sid": "MZxxxxxxxxxxxxxxxxxxxxxxxx",
  "mark": { "name": "my-label" }
}

`stop`

{
  "event": "stop",
  "sequence_number": "20",
  "stream_sid": "MZxxxxxxxxxxxxxxxxxxxxxxxx",
  "stop": {
    "call_sid": "CAxxxxxxxxxxxxxxxxxxxxxxxx",
    "account_sid": "ACxxxxxxxxxxxxxxxxxxxxxxxx",
    "reason": "callended"
  }
}

reason: stopped (applet ended) · callended (caller hung up)

Events — your server → Exotel (VoiceBot Applet only)

note

These events only apply to the VoiceBot Applet (bidirectional). The Stream Applet is receive-only — your server cannot send any events back.

`media` — send audio to caller

{
  "event": "media",
  "stream_sid": "MZxxxxxxxxxxxxxxxxxxxxxxxx",
  "media": { "payload": "<base64-encoded PCM>" }
}

`mark` — tag a playback position

{
  "event": "mark",
  "stream_sid": "MZxxxxxxxxxxxxxxxxxxxxxxxx",
  "mark": { "name": "turn-3-end" }
}

`clear` — flush buffered audio (barge-in)

{ "event": "clear", "stream_sid": "MZxxxxxxxxxxxxxxxxxxxxxxxx" }

Audio format

Property	Value
Codec	Raw PCM (linear16) — uncompressed
Bit depth	16-bit signed, little-endian
Channels	Mono
Default sample rate	8 000 Hz
Supported rates	8 000 · 16 000 · 24 000 Hz
Transport	Base64
Chunk size	3 200–100 000 bytes, must be a multiple of 320
Max session duration	60 minutes

Passthru Applet

Place immediately after the Voicebot/Stream applet. Exotel sends a GET request with stream metadata as query parameters to your callback URL when the stream ends.

Field	Description
`callsid`	Parent call ID
`streamsid`	Stream session ID
`streamurl`	WebSocket URL used
`status`	Final stream status
`duration`	Stream duration in seconds
`recordingurl`	Recording link (if enabled)
`error`	Error detail — copy as-is for support
`disposition`	Outcome classification
`disconnectedby`	`caller` · `bot` · `system`
`detailedstatus`	Fine-grained status

Active Stream Monitoring

Check how many streams are live on your account.

curl -X GET \
  'https://<api_key>:<api_token>@api.in.exotel.com/v1/accounts/<account_sid>/activestreams'

{
  "status": "success",
  "active_streams": 12,
  "max_allowed_streams": 100,
  "account_sid": "<account_sid>"
}

WSS error codes

Code	Meaning	Common cause
`1000`	Normal closure	Clean disconnect
`1001`	Endpoint going away	Server restart
`1002–1003`	Protocol / data error	Malformed frames
`1006`	Abnormal closure	Network drop · TLS failure · server crash
`1007–1009`	Payload / policy / size	Chunk out of bounds
`1011`	Server error	Unhandled exception
`1012–1013`	Restart / retry	Transient — retry

1006 checklist: WSS URL reachable → TLS cert valid → firewall allows Exotel IPs → handshake completes within 10 s.

Echo server (VoiceBot Applet / bidirectional only)

import asyncio, json, websockets

async def handle(ws):
    stream_sid = None
    async for msg in ws:
        ev = json.loads(msg)
        match ev["event"]:
            case "start":
                stream_sid = ev["start"]["stream_sid"]
            case "media":
                await ws.send(json.dumps({
                    "event": "media",
                    "stream_sid": stream_sid,
                    "media": {"payload": ev["media"]["payload"]}
                }))
            case "stop":
                break

async def main():
    async with websockets.serve(handle, "0.0.0.0", 5001):
        await asyncio.Future()

asyncio.run(main())

pip install websockets && python echo.py
ngrok http 5001  # use the https:// URL as streamurl

Stream & Voicebot Applet — Full applet and event reference
Bot Stream with Legs API — Legs API reference
Passthru Applet — Passthru configuration

How it works​

Connect Voice AI​

Parameters​

Request​

Response​

Connect Voice API with Flow​

Parameters​

Request​

Response​

Common flow patterns​

Programmable Voice APIs (ExoML)​

Step 1 — Dial the customer​

Step 2 — Start stream on leg_answered​

Optional — Play greeting while stream initialises​

Applet configuration​

Stream Applet — Unidirectional (Exotel → your server)​

VoiceBot Applet — Bidirectional (Exotel ↔ your server)​

WebSocket protocol​

Event support by applet type​

Events — Exotel → your server​

connected​

start​

media​

dtmf (VoiceBot Applet only)​

mark (VoiceBot Applet only)​

stop​

Events — your server → Exotel (VoiceBot Applet only)​

media — send audio to caller​

mark — tag a playback position​

clear — flush buffered audio (barge-in)​

Audio format​

Passthru Applet​

Active Stream Monitoring​

WSS error codes​

Echo server (VoiceBot Applet / bidirectional only)​

Related​

How it works

Connect Voice AI

Parameters

Request

Response

Connect Voice API with Flow

Parameters

Request

Response

Common flow patterns

Programmable Voice APIs (ExoML)

Step 1 — Dial the customer

Step 2 — Start stream on `leg_answered`

Optional — Play greeting while stream initialises

Applet configuration

Stream Applet — Unidirectional (Exotel → your server)

VoiceBot Applet — Bidirectional (Exotel ↔ your server)

WebSocket protocol

Event support by applet type

Events — Exotel → your server

`connected`

`start`

`media`

`dtmf` (VoiceBot Applet only)

`mark` (VoiceBot Applet only)

`stop`

Events — your server → Exotel (VoiceBot Applet only)

`media` — send audio to caller

`mark` — tag a playback position

`clear` — flush buffered audio (barge-in)

Audio format

Passthru Applet

Active Stream Monitoring

WSS error codes

Echo server (VoiceBot Applet / bidirectional only)

Related