Voice agent infrastructure

Write your agent in config.
Test it before it ships.
Own the stack.

Wax is a structured CLI and web interface for voice agents. agent.toml is the source of truth, wax deploy is the delivery mechanism, and the browser simulation lets you iterate without touching production. Built for engineers who treat voice as infrastructure, not a demo.

~/usewax 
# agent.toml — the agent definition lives here
name = "receptionist"
provider = "elevenlabs"
voice = "mimbo"
tools = ["calendar.book", "contacts.lookup"]

# deploy, run, verify
$ wax deploy --env production
Who it's for
Platform teams
You own the voice layer across product teams. Wax gives you a unified config surface and a single deploy pipeline — no bespoke per-agent scripts.
Senior engineers
You want control over the routing, the failover logic, and the cost attribution — not a black-box API you have to trust blindly.
Engineering leads
You need regression testing for voice agents, per-agent cost visibility, and tooling that fits into an existing CI/CD workflow.
The full stack, no stitching required
STT →
Voxtral, Mistral, routing through wax
LLM →
Sonnet 4.6 via Anthropic, routed through wax
TTS →
Your voice clone, ElevenLabs, bring your own key
What you get

The tooling that makes voice agents tractable at scale.

Config-driven

Agent Editor

A visual builder mapped 1:1 to agent.toml. Every field maps to a config key — no abstractions, no data loss on export.

Cross-provider failover

Define your failover routing in config. ElevenLabs, Cartesia, Deepgram — you pick the priority order and the retry policy. No proprietary magic.

Live browser simulation

Talk to your agent in the browser. A turn-card timeline streams in real-time. Reset and replay any session — no production calls needed for iteration.

Fixture-based regression testing

Run your agent against a suite of fixture conversations. Catch regressions before they hit production. Part of Builder and above.

Builder+

MCP server access

Expose typed tools to your agent. MCP server runs on your infrastructure. The agent calls tools you've defined, not whatever the provider decides to surface.

Cost attribution

Per-call, per-workspace, per-agent, per-provider breakdown. Time-series charts and breakdown tables. Know what every voice interaction costs and why.

Pricing

Pay for the management layer.
Pick how you handle provider costs.

Free
$0/mo
1,000 mins/mo
  • 1 agent
  • 1 provider
  • Voice clone
  • Basic dashboard
Builder
$299/mo
50,000 mins/mo
  • 5 agents
  • All 5 providers
  • Provider failover
  • MCP server access
  • Fixture-based regression testing
Scale
$1,499/mo
250,000 mins/mo
  • Unlimited agents
  • All 5 providers
  • SLA + priority support
  • SOC 2 Type II in progress
Most customers use prepaid credits with auto-reload — set a baseline, top up automatically, one bill, total transparency. Or bring your own provider keys.

Voice agents in production aren't demos. They're infrastructure. Wax gives you a config-first model, regression testing, and per-agent cost visibility — so you're not guessing what's running or what it's costing.