aywa runtime
Docs Sign in Create account

Private voice runtime · your infrastructure · your providers

Your voice agents, on your infrastructure. $0.019 a minute.

Measured on real production traffic — GPT-5.4 included, not a stripped-down demo config. At 20,000 minutes a month that is about $380 in provider cost plus a $49 Runtime Slot licence. Import your Vapi agents and read the itemized cost before you sign up for anything.

Make a call — no account Try the 60-minute sandbox
150 ms first audio, from end of speech $49 / month per runtime, no per-minute fee GPT-5.4 included measured on real traffic, not a cheap demo stack

The public test and shared sandbox run on Aywa infrastructure for evaluation. Your production runtime runs on infrastructure you control.

One private runtime connects voice entrypoints to the providers and data plane you control.

Live runtime test

Make a real browser call. Read its usage and cost estimate after hang-up.

No account and no installation. Choose a realtime voice-to-voice path or a cascade path, allow microphone access, and inspect the estimated provider cost, duration, pipeline, and first-audio latency from end of speech.

Inspect the metered dataset

This public test runs on Aywa infrastructure for evaluation. It is separate from the private runtime you install for production.

Public WebRTC line ready
Choose a route Realtime V2V or Cascade STT · LLM · TTS
Estimated provider cost
Metered after call
Stage breakdown
Shown for cascade
First audio from end of speech
Runtime p50 report
Account required
No

Metered evidence

We built the runtime we wanted to operate.

We wanted direct control of the infrastructure, providers, and cost structure. We imported our agents instead of rebuilding them, then measured the runtime on our own traffic.

Metered in Aywa Runtime $0.019 / min
Leading models observed
Deepgram Nova-3 · GPT-5.4 / GPT-5.4 Mini · Cartesia Sonic 3.5 / ElevenLabs Turbo v2.5
Calls
409
Ended calls
389
Minutes
908 min
Provider cost
$17.62
Cost / min
$0.019
Cost / call p95
$0.15

$0.019 is what we pay. You can go lower.

That average comes from real traffic running premium models. Swap in a faster, cheaper model and the number drops. The runtime does not change.

You pick the trade-off. We just stop charging you a fee on top of it.

Cost per call at p95: 3.3× the average.

A tight tail is what makes a call centre budgetable. Most platforms publish an average and let you find the outliers on the invoice.

We took a hosted invoice apart line by line

Estimated tracked provider cost from metered usage and configured or public rate cards.

Valued at configured or public provider rate cards. Excludes carrier, infrastructure, storage, taxes, the $49 Runtime Slot, provider credits, and committed-use discounts.

Use your own numbers

What this costs at your volume.

Every field is editable. Aywa Standard adds a $49 monthly licence; provider usage and infrastructure stay in your own accounts.

Recommended server 4 vCPU / 16 GB / 200 GB NVMe
Estimated infrastructure $13–$29 / month based on 24-month committed VPS pricing
Provider usage
$380
Runtime licence
$49
Infrastructure Editable — the reference high end is used by default
Your monthly cost
$458

Provider usage is billed to your own accounts. Carrier, storage, taxes and engineering time are not included on either line. The concurrency estimate assumes 22 business days, 8 calling hours a day, and a 2× peak factor.

Capacity planning

What server do I need?

Concurrency is what sizes a voice runtime, not monthly volume. A hundred thousand minutes spread thin needs less machine than ten thousand minutes that all land at 9am.

Recommended operating envelope: 50 concurrent calls. Not a licence limit. Load-tested with 150 simultaneous voice calls on a 4-vCPU, 16-GB machine. The recommendation preserves operating headroom for traffic bursts, provider mix, and call profile. The Standard licence does not impose a concurrency cap or add a per-minute Aywa fee.

One runtime handles this comfortably. That's not why you'd want a second one.

You'd want a second one because the first one can reboot.

  • Single point of failure

    A single machine is a single point of failure.

  • Live updates

    It can't be updated without dropping live calls.

  • Regional latency

    It serves one region well and distant regions less well.

  • Blast-radius isolation

    It puts every client's traffic in the same blast radius.

Peak concurrent calls Recommended server Estimated
1–50 4 vCPU / 16 GB / 200 GB NVMe $13–$29 / mo
Above 50 or when uptime matters Multiple runtimes — see Enterprise Custom

Measured on our own runtime. Reference pricing checked August 2026, based on 24-month committed VPS rates. Your provider mix and call profile will shift these numbers — the calculator above takes your own figures. Pricing reference.

Three ways to see it work

Start with the amount of proof you need.

Each step answers a different question. You do not need to provision a server before hearing a real call.

01 · 90 seconds

Live call

Answer “does the runtime work?” with a real WebRTC call and an itemized post-call cost estimate.

  • No account
  • No provider keys
  • Realtime or cascade
02 · Self-serve

Shared sandbox

Build and call an assistant in an isolated evaluation organization before installing anything.

  • 60 cumulative Runtime minutes
  • Your own provider API keys
  • WebRTC, evaluation only
Open the sandbox
03 · Your infrastructure

Private 14-day install

Validate deployment, migration, providers, call quality, and operations on a VPS you control.

  • One private Runtime Slot
  • Docker Compose installer
  • No credit card for qualified technical evaluations
Install it on my box

Sandbox means hosted evaluation. Standard means a licensed runtime installed on your infrastructure. Aywa does not sell hosted production calls.

Keep the work you already did

Already on Vapi? Keep the agents you already built.

Import assistants, tools, squads, phone numbers, webhooks, files, and structured outputs. Reconnect your provider keys, validate representative calls, then cut over. No rebuild from zero.

See the migration path Review the import scope

Imports preserve compatible configuration, not guaranteed behavioural parity. API keys are reconnected rather than imported.

  1. 01
    Preview

    Inspect assistants, tools, squads, phone numbers, webhooks, files, structured outputs, warnings, and source-ID mappings.

  2. 02
    Import compatible resources

    Preserve prompts, definitions, bindings, and linked references where the source and runtime contracts match.

  3. 03
    Reconnect and test

    API keys are not imported. Reconnect provider, SIP, and storage settings, then validate representative calls before traffic moves.

What you get

The production operating layer—not another orchestration sketch.

Aywa Runtime packages the real-time loop, provider boundary, call diagnostics, and release operations into one supported runtime.

01

Run the human turn

Streaming STT, endpointing, model first-delta tracking, segmented TTS, barge-in, media playout, and realtime voice-to-voice.

Measure latency stage by stage on representative calls.
02

Bring your provider stack

Connect supported STT, LLM, TTS, telephony, storage, and analytics accounts with your own keys.

Providers bill you directly; Aywa adds no markup to provider minutes.
03

Debug from inside the call

Inspect provider timing, endpointing, tool attempts, signed webhooks, retries, warnings, metrics, and artifacts by call ID.

Share redacted support bundles without exposing raw audio or secrets by default.
04

Ship controlled updates

Use signed images, a stable release channel, preflight checks, readiness verification, backups, and documented rollback.

The control plane manages releases; production call execution remains in your runtime.
OpenAIAnthropicDeepgramElevenLabs CartesiaTelnyxTwilioPostgres See supported adapters →

Support and operations

Self-hosted doesn’t mean self-supported.

You run Aywa Runtime inside your infrastructure. You do not debug the Aywa software alone. Standard email support covers the runtime, installer, activation, registry and API compatibility, stable updates, and reproducible runtime issues.

Start with Aywa support Read the support boundary

Standard support covers the Aywa software boundary. VPS/cloud accounts, network policy, provider contracts, DNS, TLS, backups, and incident ownership remain with the operator unless Enterprise scopes them explicitly.

Direct technical support A direct support path, not a community-thread dependency.

Open a support ticket with reproducible context and a customer-approved, redacted diagnostic bundle.

DP Documented deployment

Installer, preflight checks, readiness verification, backup guidance, rollback, and documented import tooling.

UP Controlled updates

Stable release channel, preflight checks, readiness verification, documented backups, and rollback.

DX Call-level diagnosis

Timelines, provider latency, tool attempts, webhook retries, warnings, metrics, and artifacts by call ID.

RB Safe support bundles

Runtime context and dependency status without provider secrets, raw audio, or full transcripts by default.

Built and supported by AYWA

A product with a named operator and inspectable technical surfaces.

Aywa Runtime is a product of AYWA, not a second company. Tariq Lambachri is Founder @ AYWA and is directly accountable for the product, its release path, and its technical support boundary.

We do not display invented customer logos or anonymous scale claims. Evaluate the live instrument, public documentation, SDK package, security model, and the runtime on your own infrastructure.

Product owner Tariq Lambachri

Founder @ AYWA

Public package@aywa-ai/runtime-sdkApache-2.0 SDK on npm OperationsSupport and updatesRelease, rollback, and support boundary Deployment boundarySecurity modelInspect data flow and operator responsibilities

Private runtime pricing

One production runtime slot. Enterprise when you need more.

Standard is a single licensed private runtime installed on your infrastructure. Aywa charges the runtime license; STT, LLM, TTS, telephony, storage, and analytics stay with your provider accounts, so usage is paid at source with no Aywa platform markup on call minutes.

One Runtime Slot = one active private Aywa Runtime deployment.

Runtime license prices exclude applicable taxes. Stripe applies configured VAT or local tax registrations at checkout from the billing address and business tax ID when provided.

Additional Runtime Slots, volume pricing, HA, load balancing, private builds, and accompanied installs are Enterprise custom.

Trial
$0 / 14 days

No credit card required for qualified technical evaluations.

One Runtime Slot to validate deployment, import flow, provider credentials, voice-to-voice calls, and call quality.

  • 1 private Runtime Slot
  • Installer and Docker Compose
  • Trial license token
Start trial
Standard
$49 / mo

One production Runtime Slot. $49/month or $490/year, before applicable taxes.

For one production deployment with the runtime dashboard on customer infrastructure and provider costs paid directly.

  • 1 production Runtime Slot
  • Stable release channel
  • Runtime support and redacted diagnostics
  • Direct provider billing with no Aywa minute markup
Choose Standard
Enterprise
Custom

Volume pricing for multiple Runtime Slots.

For teams that need multi-runtime, HA, load balancing, private builds, SLA, or installation led by Aywa.

  • Custom Runtime Slots and groups
  • Managed Edge / HA architecture
  • Guided deployment and agreed SLA
  • Private release policies
Contact sales

Before you choose self-hosting

Four boundaries worth making explicit.

Aywa Runtime gives you control, but it does not make infrastructure or provider operations disappear.

Is self-hosting always cheaper?

No. Your volume, provider mix, telephony, infrastructure, storage, and operating time determine the result. Use the calculator above with your own bills, then validate representative calls before deciding.

Where does Aywa support stop?

Standard email support covers the Aywa Runtime software, installer, activation, registry and API compatibility, stable updates, and reproducible runtime issues. Your cloud or VPS, network, provider contracts, DNS, TLS, backups, and incident ownership remain your responsibility unless an Enterprise agreement scopes them explicitly.

What happens if the Aywa control plane is temporarily unavailable?

Active calls are not interrupted. Current Standard defaults cache a signed lease for up to 24 hours and allow up to seven additional days of grace before new calls are blocked; active calls still drain. This is a bounded resilience mechanism, not a promise of perpetual offline use.

What happens to my data and configuration if I leave Aywa?

Runtime data and configuration remain on infrastructure you control. Exporting or moving to another system still requires compatibility checks and validation; we do not claim automatic behavioural parity.

Enterprise path

When one runtime slot is not enough, we design the deployment with you.

Standard stays intentionally simple: one licensed runtime, installed by the customer. Enterprise is for teams that need multiple runtimes, private release policies, HA/edge design, procurement, security review, or an accompanied production cutover.

Contact sales Start Standard first
Fit criteria

Use Enterprise for architecture, not just volume.

We scope runtime topology, rollout gates, provider routing, support boundaries, and operational ownership before any custom contract.

HA HA and edge design

Customer-owned load balancer, SIP/media edge choices, failover expectations, and runbook ownership.

PB Private builds

Pinned releases, private channels, rollout windows, and signed images for controlled environments.

MS Migration support

Import mapping, provider credential plan, phone-number cutover, and validation calls before traffic.

SR Security review

Data boundary, support bundle policy, access model, audit events, and deployment evidence package.