API layer access · Alebex
Partners / API layer access

Take access at the API layer

Voice sessions, streaming and call control through open REST and WebSocket APIs.

The engine is reachable three ways
REST

Place and manage calls, configure agents, pull transcripts.

WebSocket

Audio streaming for live sessions.

CLI

Test calls, session replays, experiments.

REST, WebSocket and a CLI, against the same runtime.

On this page What the API reaches into The integration What integrating costs Four innovations What partners build
Partners

What the API reaches into

Quality rides alongside the stream — never in series with it.

Telephony
8 kHz mulaw over media streams
Browser voice
16 kHz PCM

One pipeline serves both, with zero code forks.

Alebex Voice Engine — the stream, in order
01
Audio in
02
Noise reduction
Per frame
03
Voice activity detection
04
Streaming speech-to-text
05
Turn control
Turn end + barge-in
06
LLM
Token stream, gated
07
Streaming text-to-speech
Per sentence
08
Audio out
Interruptible

Alebex Voice Engine — the stream, in order.

Developers

What the integration looks like

REST opens the session and places the call; the socket carries the audio. Both shapes below are the ones the platform itself is built on.

Streaming session cURLTypeScriptPython Copy Illustrative 
# 1 · Open a streaming session over REST.curl https://api.alebex.ai/v1/speech-to-text/sessions \  --header "Authorization: Bearer $ALEBEX_API_KEY" \  --header "Content-Type: application/json" \  --data '{    "model": "alebex-asr",    "audio": { "encoding": "pcm_s16le", "sample_rate": 16000 },    "interim_results": true  }' # {#   "id": "ses_...",#   "socket_url": "wss://api.alebex.ai/v1/speech-to-text/stream?session=ses_..."# } # 2 · Open socket_url, send audio frames as they arrive, read transcripts back.# You are never waiting on a request to return. # 3 · Pull the finished transcript over REST when the session closes.curl https://api.alebex.ai/v1/speech-to-text/sessions/ses_.../transcript \  --header "Authorization: Bearer $ALEBEX_API_KEY"

Shell request with cURL. Illustrative sample. Endpoint paths, parameter names and field names are confirmed in the documentation issued with your account.

REST Copy Illustrative 
# REST places and manages calls.# 555-01xx is the reserved fictional range.curl https://api.alebex.ai/v1/calls \  --header "Authorization: Bearer $ALEBEX_API_KEY" \  --header "Content-Type: application/json" \  --data '{ "to": "+16045550142", "agent": "agt_..." }' # Pull the transcript once the call has ended.curl https://api.alebex.ai/v1/calls/cal_.../transcript \  --header "Authorization: Bearer $ALEBEX_API_KEY"

Shell request with cURL. Illustrative sample. Endpoint paths, parameter names and field names are confirmed in the documentation issued with your account.

Developer workstation with two monitors of code, night, cyan monitor glow
Integration effort

What integrating actually costs you

The cost is effort: the boundary between the application you own and the runtime you do not have to assemble.

You build
Your application and workflow

Build the client-facing application, connect your backend and define the workflow the voice session serves.

You skip
A second voice stack

Do not assemble streaming speech-to-text, turn control, the hallucination gate, barge-in classification, multimodal intent and streaming synthesis from separate layers.

You operate
The client-facing boundary

Operate your application, credentials, client workflow and first-line support.

Alebex operates
The runtime underneath

Alebex operates the voice runtime and Canadian data residency. Recordings, transcripts and customer information are held in Canada, with PIPEDA and CASL compliance built in.

Alebex Voice Engine

Four documented innovations

Each one is a layer the engine owns rather than assembles, which is why it can sit inside the stream instead of in front of it.

01 Context-aware linear streaming speech-to-text Full-context transcription at linear cost: re-decoding’s coherence at chunking’s price, any turn length.
02 The parallel hallucination gate Latent-space check plus NLU confirmation riding the synthesis window, verdict in under 50 ms. under 50 ms Anti-hallucination verdict
03 Barge-in classification A learned classifier on raw audio (WavLM-based) that separates real interruptions from backchannels and line noise in milliseconds.
04 Multimodal intent understanding An LLM decoder that consumes audio and text together and writes the intent label: open vocabulary, dynamic per deployment, capturing the hesitation and tone a transcript flattens.
Alebex Voice Engine
Partners

What partners build on it

A developer ecosystem building custom applications on the platform, with self-serve developer accounts and the models callable directly over the API.

API access →

Voice sessions, streaming and call control through open REST and WebSocket APIs.
Developer accounts →

Self-serve accounts, and the models callable directly.
Alebex Voice Engine →

The runtime everything streams through.
Developers

Build with our API

Voice sessions, streaming and call control through open REST and WebSocket APIs.