Everything between the
ring and the hang-up
Aura owns the entire stack — perception, reasoning and action — so your agents are fast, grounded, and able to actually get things done on a call.
A voice pipeline tuned to the millisecond
Streaming everything. The caller hears the first syllable of the reply before the model has finished generating it — that's where the sub-600ms feel comes from.
Voice activity detection
Silero VAD decides the exact instant a caller stops speaking — no awkward two-second wait.
Streaming STT & TTS
Deepgram interim transcripts in, ElevenLabs μ-law audio out — both streamed, never buffered to completion.
Barge-in & backchannels
Callers can interrupt naturally; the bot drops what it was saying. "Mm-hmm" backchannels signal it's still listening.
Personas that know your business and act on it
Above the raw STT-LLM-TTS plumbing sits the Brain: persona, intent, retrieval, tools, memory and workflows working as one.
Configurable personas
Each agent has a system prompt, greeting, voice, language and allowed tools — version-managed, hot-editable.
RAG over pgvector
Per-agent and tenant-wide knowledge with cosine top-k retrieval, speculative pre-fetch and Redis caching for follow-ups.
Tools with human approval
Agents call your APIs mid-call. Sensitive actions queue an approval card for one-click accept/reject in the dashboard.
Caller & within-call memory
Cross-call profiles greet returning callers warmly; rolling summaries keep long calls inside the context window.
More than 30 capabilities, production-ready
Each is battle-tested in live, high-volume production deployments.
Inbound & outbound
Accept calls on assigned numbers routed by DID, or fire outbound campaigns over REST — same persona, same pipeline.
Live transcripts
Every turn streams to the dashboard over SSE/WebSocket. Watch, flag and QA calls as they happen.
Post-call summaries
Automatic summaries, sentiment and outcomes generated the moment a call ends — ready for your CRM.
40+ languages
Mid-call code-switch detection hot-swaps STT, LLM and TTS to the caller's language without dropping the thread.
Custom voice clones
Upload reference samples to clone a brand voice, or pick from a catalog of provider voices per agent.
Smart routing & fallback
The router picks the best LLM/STT/TTS per turn by capability, cost and health — circuit-breaking to fallbacks mid-call.
Warm escalation
When turn counts cross a threshold, agents warm-transfer the live call to a human escalation number.
Automated voice tests
Define test suites and replay scenarios to catch regressions before they reach a real caller.
Embeddable web widget
Drop a script tag for browser-side WebRTC calls — let visitors talk to an agent right from your site.
Enterprise security, multi-tenant by design
Field-level encryption
Twilio credentials and tool secrets are Fernet-encrypted at rest.
Three auth schemes
JWT users, scoped X-API-Key integrators, and validated Twilio signatures.
GDPR export & erase
One-click data export and right-to-erasure, with an append-only audit trail.
Observability
Prometheus metrics, OpenTelemetry traces and Sentry with PII scrubbing.
See the platform on a real call
We'll build an agent on your knowledge base and call you live in the demo.