Documentation for the people who will run it.

Everything you need to build, deploy, and operate Wordcab — private voice AI that stays inside your boundary. Guides, API reference, SDKs, CLI, and operator runbooks.

Start here

Wordcab is private voice AI infrastructure. The same stack that powers cloud API traffic is what ships into your VPC, your datacenter, or an airgapped environment. These docs cover both modes — the public API at api.wordcab.com and the self-hosted runtime you install with Helm or the operator.

Get started

Quickstart

Make your first request in under five minutes — transcription, speech, and a chat completion.

 Auth

Authentication

API keys, scopes, rotation, and the pattern for short-lived tokens on self-hosted deployments.

 Build

Build a voice agent

System prompt, voice selection, tool use, and a first outbound call — end to end.

 Migrate

OpenAI compatibility

Point your existing OpenAI SDK at Wordcab and keep your application code unchanged.

What you can build

The Wordcab API groups into three product surfaces. They share authentication, billing, control plane, and deployment artifacts — but you can adopt them independently.

Voice

Transcription & speech

Streaming and batch STT (Qwen3-ASR, Voxtral Realtime, Cohere Transcribe 2B). Streaming TTS (Qwen3-TTS, Kokoro).

 Think

LLM inference & reasoning

Chat completions, embeddings, tool use, JSON mode. Gemma 4, Qwen3.5, DeepSeek V3.2, Llama 3.3.

 Adapt

Evaluation & fine-tuning

Prepare data, run held-out evals, fine-tune against your real audio, and validate before rollout.

Developer surfaces

Three equivalent ways to talk to the Wordcab control plane. Pick whichever fits the task.

HTTP

REST API

OpenAPI 3.1, Bearer-token auth, JSON everywhere. OpenAI-compatible /v1 endpoints for chat, embeddings, and audio.

 SDK

Python & TypeScript SDKs

Typed clients with retries, pagination, and streaming helpers baked in. Drop-in replacement for the OpenAI SDK when you want it.

 CLI

Command line

Scriptable wordcab CLI for transcription, speech, agents, deployments, and log streaming against any environment.

Deploy & operate

When you run Wordcab inside your own infrastructure — VPC, on-prem Kubernetes, airgap, or hybrid — the same API runs behind a Helm-installed control plane. Everything below is operator-facing.

Start here

Self-hosted overview

Deployment shapes, reference hardware, what ships with the chart, time to first call.

 Install

Helm chart

Prerequisites, values.yaml, operator CRDs. wordcab deploy apply wraps it with preflight.

 Distros

Kubernetes

EKS, AKS, GKE, OpenShift, RKE2 — per-distribution notes on ingress, storage, GPU operator.

 Offline

Airgap installs

Signed bundles, Cosign verification, internal registry import, preflight.

 Day 2

Upgrades & rollback

Rolling upgrades, one-command rollback, cadence, and stability rules.

 Observe

Observability

Prometheus, OpenTelemetry, structured logs. Six Grafana dashboards and an SLO alert pack ship in the chart.

 Auth

Identity & SSO

SAML, OIDC, SCIM, workload identity, audit-to-SIEM. Configured at install time.

 Voice

Telephony & SIP

Twilio media streams, native SIP for on-prem PBX, Genesys / Five9 / Zoom connectors.

 Frameworks

Framework integrations

Pipecat, LiveKit, Daily, Vapi, Retell, LangChain, LlamaIndex.

 Backends

Model serving

vLLM (default), SGLang, Triton, ONNX Runtime. Backend choice is per pool.

 Architecture

Deployment shapes

Reference diagrams for VPC, on-prem, airgap, and hybrid. Context for the operator docs above.

 Control

Deployments API

Programmatic management of environments, routing, and autoscaling.
Access during evaluation

Some pages — security review bundles, DPA/BAA templates, offline bundle contents — are shared under NDA during Pilot. Request docs access to get the full bundle.