Fit the stack your team already runs.
Wordcab plugs into the orchestration, telephony, model-serving, and infrastructure layers enterprise voice teams already run. OpenAI-compatible endpoints. Standard Kubernetes packaging. Open telemetry. Every major private-cloud target.
Voice agent orchestration
Drop Wordcab Voice and Think into the real-time agent frameworks your team already builds on. OpenAI-compatible endpoints — existing code works without changes.
Wordcab Voice ships as a Pipecat service for STT, TTS, and LLM. Tested against Pipecat 1.0 (Apr 2026). Full async pipeline support with VAD, turn detection, and interruption handling.
Drop-in provider plugin for LiveKit Agents. Works with LiveKit Cloud or a self-hosted SFU. Audio stays inside your LiveKit instance — no round-trip to a hosted STT vendor.
Use via Pipecat's Daily transport or directly against the Daily Bots API. Common pattern for meeting capture and bot-assisted workflows.
Bridge Twilio Media Streams (μ-law / 8 kHz) straight into Wordcab Voice via WebSocket. Reference app ships with full TwiML setup and a Pipecat-compatible transport.
Use Wordcab as a custom STT/LLM provider inside Vapi agents. OpenAI-compatible chat endpoint plugs in without custom glue.
Custom LLM URL pointed at a Wordcab Think deployment. Reasoning stays inside your boundary while Retell handles telephony glue.
Telephony & contact center
SIP-native ingress, CCaaS connectors, and reference adapters for the platforms where contact center audio actually lives. Full-call coverage on telephony-grade codecs.
Direct SIP ingress via an Asterisk- or FreeSWITCH-backed gateway. μ-law / A-law / Opus. Reference chart ships with the Helm package.
AudioHook protocol adapter streams Genesys call audio into Wordcab Voice for real-time transcription, QA, and redaction.
VoiceStream API adapter. Transcripts and QA signals delivered via webhook or Kafka back to Five9 reporting.
Real-time Audio Streaming connector in active development with a design partner. Q3 2026 target.
Media stream access via Zoom's RTMS / real-time media gateway for business-communications workflows.
Same WebSocket media stream model as Twilio. One adapter covers programmable-voice vendors.
Model serving & inference runtimes
Wordcab Voice and Think run on the serving stack your platform team trusts. OpenAI-compatible endpoints — application code does not change when models do.
Default serving backend for Wordcab Think. Qwen3.5, Gemma 4, and Cohere Transcribe all ship with vLLM-tuned configurations and tensor-parallel presets.
For structured-output and multi-turn agent workloads where SGLang's constrained decoding and radix cache win on throughput.
TensorRT-LLM and ONNX Runtime backends for teams standardized on Triton. Model repository layout and config files included.
OpenAI-compatible /v1/chat/completions, /v1/embeddings, /v1/audio/transcriptions, and /v1/audio/speech. Point your existing SDK at Wordcab’s base URL—same shapes, runs inside your boundary.
CPU and edge-targeted deployments. Kokoro TTS and smaller Think models run at latency competitive with GPU at low concurrency.
Deployed with SCX.ai for Australian sovereign inference. Sub-100 ms ASR latency and meaningfully lower power-per-inference than equivalent GPU.
Infrastructure & operations
Fits the tooling your platform engineering team already standardized on. Terraform, Helm, and OpenTelemetry — not custom vendor consoles.
Production Helm charts with values-based environment overrides. Tested on EKS, GKE, AKS, OpenShift 4.x, and RKE2. Operator available for advanced lifecycle management.
AWS, Azure, and GCP modules provision VPC, GPU node pools, and managed dependencies. Drop into existing Terragrunt or Spacelift workflows.
Every service emits Prometheus metrics and OpenTelemetry traces. Grafana dashboards ship with the chart. No proprietary telemetry path.
Okta, Azure AD, Google Workspace, Auth0, Keycloak. SCIM provisioning for enterprise directories. Audit logs streamable via webhook or syslog.
Private registry mirror, signed offline bundles, preflight checks, custom CA chains. No hidden call-home in the critical path.
Transcripts, summaries, redaction events, and QA signals delivered via HTTP webhooks or Kafka topics. At-least-once delivery with replay.
SCX.ai — sovereign AI infrastructure
Wordcab Voice runs on SCX.ai's SambaNova-RDU-backed sovereign infrastructure in Australia. The joint deployment delivers private voice AI for financial-services and public-sector customers whose processing must stay inside Australian jurisdiction — with the power-per-inference economics of purpose-built accelerators.
Engagement model: co-deployment, shared engineering on custom voice models, and a public provider surface at sam.scx.ai. A reference for any team evaluating sovereign or RDU-based inference for voice.
Missing the integration your team needs?
Tell us the orchestration framework, CCaaS platform, or model runtime you need Wordcab to fit into. If it is not on this page, it is usually a two-week engineering effort.
Talk to an Engineer
We usually respond within one business day.