Private voice AI infrastructure
Deploy production AI phone agents on your own VPS, cloud, or Kubernetes cluster. Keep your providers, data flow, call capacity, and runtime updates under your control. Calls, transcripts, recordings, and provider secrets stay inside the infrastructure you deploy.
$ aywa deploy --target vps-01
runtime registered
license active
voice pipeline ready
clickhouse sink connected
sip edge reachableRuntime integrations
The provider grid reflects what the runtime can actually route today: model adapters, OpenAI-compatible LLM routes, streaming STT, telephony-ready TTS, carrier/SIP/WebRTC, and the production data plane.
Voice experience engine
The runtime is built around the real-time phone experience: deciding when the user has finished speaking, streaming the model early, keeping TTS playable on telephony, and measuring latency across every stage.
Semantic turn classifier, duplicate final suppression, line-check filtering, assistant-active deferral, and fast opening/post-opening paths.
Telephony audio normalization, loudness leveling, soft limiter, speed adjustment, TTS continuity, and a single-output call mixer.
STT preconnect, streaming LLM first-delta tracking, segmented TTS playout, provider retries, fallback voices, and runtime SLO metrics.
Function tools, native call controls, signed server messages, webhook outbox retries, redacted logs, and outbound request allowlists.
Why private runtime
Hosted platforms are excellent for prototypes and managed launch speed. Aywa Runtime is built for teams that are ready to own the execution layer: cost model, capacity planning, provider routing, privacy boundary, and production debugging.
We describe infrastructure tradeoffs as a product category. Aywa Runtime is independently developed and does not imply affiliation with any third-party voice platform.
License the runtime separately and keep STT, LLM, TTS, telephony, storage, and analytics billing with the providers you choose.
Scale on your VPS, cloud, or Kubernetes design instead of treating every concurrency decision as a hosted-plan constraint.
Audio, transcripts, recordings, tool payloads, provider secrets, and runtime logs stay in the infrastructure you operate.
Inspect calls from inside the runtime: provider timing, endpointing, tool attempts, webhook retries, and redacted support bundles.
Architecture
The SaaS manages licensing, releases, instance health, and migration workflows. The call path, provider secrets, recordings, and telephony edge stay in the customer deployment.
STT, LLM streaming, TTS, barge-in, tools, webhooks, retries, and call state stay in process.
Data residency and GDPR posture
The call path, audio, transcripts, recordings, provider credentials, and webhook payloads are processed by the runtime you deploy. Aywa Control Plane only needs licensing, release, health, and migration metadata.
aywa runtime helps teams operate with data minimization, private deployment, retention controls, and redacted support workflows. It does not automatically certify a customer deployment as GDPR compliant.
Run the runtime, database, object storage, analytics, and SIP edge in your chosen region or VPS.
The control plane does not sit in the media path for phone calls, recordings, or transcripts.
OpenAI, Deepgram, ElevenLabs, SIP, storage, and webhook secrets belong to the customer deployment.
Use your own Postgres, ClickHouse, and S3-compatible storage policies for logs and artifacts.
Deployment
Start on a VPS, then graduate to multi-node deployments with Postgres documents, Redis leases, S3-compatible artifacts, ClickHouse analytics, and controlled release channels.
curl -fsSL https://install.aywaruntime.com | \
sudo AYWA_TOKEN="$AYWA_INSTALL_TOKEN" bash
aywa status
aywa update --channel stable
aywa support-bundle --redactMigration center
Preview imports, map source IDs to runtime IDs, reconnect redacted credentials, and convert imported resources into Aywa's own runtime schema.
Third-party product names are used only to identify supported import sources. Aywa is independent and is not affiliated with, sponsored by, or endorsed by those third parties.
Use a short-lived source API key for a preview or import request.
Inspect assistants, tools, credentials, files, numbers, webhooks, and ID remaps before writing.
Reconnect secrets, attach providers, and place calls through the private deployment.
Private runtime pricing
License the runtime, run it where you want, and keep provider billing between you and your providers. No Aywa platform markup is added to call minutes.
Every paid plan includes signed runtime images, signed licenses, and private registry access while the subscription is active. Trial licenses are time-limited and include one runtime instance.
No credit card required for qualified technical evaluations.
One runtime instance to validate deployment, import flow, provider credentials, and call quality.
Billed monthly for teams validating private runtime operations.
For dev, staging, and small production deployments that need a private runtime.
Billed monthly when you want production without annual approval friction.
Two private runtime instances for production teams that want predictable infra pricing.
Billed monthly for agencies standardizing private deployments.
For teams managing multiple client deployments with repeatable templates and support needs.
Monthly or annual terms available for larger deployments.
For regulated teams that need onboarding, SLA, audit, private builds, or custom hardening.