Breeze Buddy
Voice AI Platform
Breeze Buddy is an enterprise-grade conversational voice AI platform with fully-managed infrastructure — the system wrapped around the model. Deploy intelligent voice agents at scale across phone calls and web applications, without writing agent logic or running the operational stack yourself.
What is Breeze Buddy?
Breeze Buddy is AI + Infra. You get the intelligence (LLM-driven flows, real-time STT/TTS) and the infrastructure that keeps it online (isolated per-call runtime, pools, retries, storage, observability) — one platform, both layers.
It handles the entire lifecycle of a voice conversation — receiving a lead, placing or connecting a call, running a real-time speech pipeline (STT → LLM → TTS), navigating a template-driven flow, and delivering structured outcomes via webhooks.
Think of it as a programmable voice agent as a service: you define what the agent says through JSON templates, and Breeze Buddy handles the how.
Define conversation flows, prompts, and transitions in JSON templates. Deploy changes instantly without code releases.
Use Daily WebRTC for browser-based voice or Telephony (Twilio, Plivo, Exotel) for phone calls — same template, same API.
1-pod-1-call runtime via Agent's Cloud — dedicated Kubernetes pod allocated per call, media stream pinned to the pod, automatic horizontal scale-out. We run the cluster, you don't.
Multi-tenant RBAC, blacklists, analytics, Langfuse auto-evaluation + Slack alerting, OpenTelemetry tracing — quality gates and compliance built in.
How is it different?
Most voice AI tools give you an LLM wrapper and leave the operational stack to you. Breeze Buddy ships both:
| Capability | Typical Voice AI | Breeze Buddy |
|---|---|---|
| Agent logic | Custom Python/Node code | JSON templates — no deployment needed |
| Conversation flow | Hard-coded state machines | Graph-based flow nodes with LLM-driven transitions |
| Voice pipeline | Stitch together STT + LLM + TTS | Managed pipeline — Pipecat under the hood |
| Call isolation | Shared workers, noisy-neighbour risk | One call, one pod — Agent’s Cloud allocates a dedicated pod per call and releases it on end |
| Session startup | ~8–10s cold-start | ~3–4s via pre-warmed process + room pools |
| Telephony | Write Twilio/Plivo integrations | Built-in — three providers, auto-retry, warm transfer |
| Provider switching | Rewrite integration layer | Per-template config — swap STT/TTS/LLM in JSON |
| Multi-tenancy | Build yourself | Native RBAC + KMS-wrapped per-merchant credentials |
| Observability | Instrument everything | Langfuse + OpenTelemetry + auto-eval alerting built-in |
Key use cases
- Appointment reminders & confirmations — outbound calls with date/time confirmation and rescheduling
- Lead qualification — automated voice screening with structured data collection
- Customer support IVR — inbound call handling with intent routing and warm transfer to humans
- Survey & feedback collection — multi-question voice surveys with sentiment analysis
- In-app voice assistants — embed real-time voice AI in web/mobile apps via Daily WebRTC
Choose your integration
Breeze Buddy supports two transport modes. Both use the same template system and Leads API — the only difference is how audio reaches the user. See the Architecture page for the full request lifecycle.
Embed real-time voice AI into web and mobile apps. Low-latency audio via Daily's WebRTC infrastructure, RTVI events, and recording support.
Place and receive phone calls through Twilio, Plivo, or Exotel. Outbound campaigns, inbound routing, and warm transfer.
How it works
The core workflow is simple:
- Push a lead via API
Send a request to the Leads API with your template, payload, and execution mode
- Platform validates and routes
Pre-checks, template schema match, and number blacklist are validated. Then the lead is routed to Daily or Telephony.
- Voice pipeline processes in real-time
STT → LLM → TTS pipeline processes audio. The LLM navigates template nodes via function calling.
- Receive webhook callbacks
Get full transcription, collected data, outcome status, and analytics via the reporting webhook.
See the full picture
For the detailed request lifecycle with flow diagrams, execution modes, and platform stack, see the Architecture page.
Customize your agent
Every aspect of the voice experience is configurable through JSON templates. No code deployment needed.
Build conversation graphs with directed transitions, branching, and per-node config
Functions & HooksDefine LLM-callable functions for navigation and data collection with lifecycle hooks
STT / TTS ProvidersSwap between Soniox, Deepgram, Google, ElevenLabs, Cartesia per template
VAD & InterruptionFine-tune voice activity detection, turn-taking, and barge-in handling
Next steps
System design, request lifecycle & flow diagrams
AuthenticationSet up API tokens and RBAC
Quick StartYour first voice agent in 5 minutes
Build a TemplateDesign your conversation flow