Cashcrown // voice assistant · 10/11
Voice Concierge
voice-conciergeliveA phone voice agent grounded in your knowledge base (RAG): it answers routine calls, books appointments, and hands difficult cases to a human with full context.
self-host
deploy
zero-retention
recordings
<2s
response time
4 tryby
call paths
01
Problem
The phone rings when there is no one to answer, and half the calls are the same questions about hours, status, and booking. Missed calls are lost appointments, and staffing a desk for routine work is expensive. Off-the-shelf voice bots lose context and never know when to hand a call to a human.
02
Capabilities
- answers routine calls: status, hours, common questions
- books and reschedules appointments into free calendar slots
- answers grounded in your knowledge base (RAG), no making things up
- hands complaints, emotional, and irreversible cases to a human with full context
- no payments or account changes by voice: only with human confirmation
03
Delivery scope
- A SIP/WebRTC phone voice agent with low latency (<600ms) and natural Neural TTS voice synthesis
- Integration with a RAG knowledge base ensuring answers are grounded solely in approved procedures
- Automated booking and rescheduling in the calendar with availability checks and SMS confirmations
- Safe escalation to a human agent with call summaries and a strict prohibition of unauthorized actions
We assemble this product from measured models and components - see the full live pipeline:
04
Proof-led pilot
Pilot package: phone concierge grounded in knowledge
We deploy Voice Concierge on a dedicated phone line: the agent takes calls 24/7, answers strictly from your approved knowledge base, and books calendar slots without hallucinations.
<600ms
voice response latency
24/7
uninterrupted line availability
100%
context on human handoff
Evidence
Playground: knowledge retrieval and citation engine
See how the RAG layer queries business documentation and prevents hallucinations before audio synthesis.
view proofVoice operations security principles
Strict prohibition of financial transactions via voice and secure human handoff protocols.
view proofPilot path
- 01Step 1: Ingestion of support workflows and FAQ into a local cited knowledge index
- 02Step 2: SIP trunk configuration, voice naturalness tuning, and booking calendar integration
- 03Step 3: Controlled pilot on 10% call volume with instant human escalation for complex matters
05
Integrations
routerBGE-M3QdrantOllamafaster-whispern8n
06
Architecture
Select a node to see its description and data flow.
Architecture - text description
- Ingest / data
- Data sources: Content, documents and events feed the index. (Flows to: BGE-M3)
- BGE-M3: Local embeddings (1024-dim) - PII stays on-prem. (Flows to: OpenClaw router)
- AI core
- OpenClaw router: The only entry to the models - masks PII, picks the model, throttles. (Flows to: Voice Concierge)
- Voice Concierge: A phone voice agent grounded in your knowledge base (RAG): it answers routine calls, books appointments, and hands difficult cases to a human with full context. (Flows to: API)
- Serve / integrations
- API: REST surface - cited answers with access control. (Flows to: router, BGE-M3, Qdrant, Ollama)
- router: Integration: router.
- BGE-M3: Integration: BGE-M3.
- Qdrant: Integration: Qdrant.
- Ollama: Integration: Ollama.
07
Live demo
08
FAQ
How much does it cost and what does the rollout look like?
We work in ranges that depend on the number of lines, languages, and call paths: we start with a fixed-cost pilot on one line and a few paths, so you hear the quality before a larger investment. When the agent handles dozens of routine calls a day and rescues missed appointments, it usually pays off in 2–4 months. Calculate the return in the ROI calculator.
Does it integrate with our calendar and telephony, and is it GDPR-compliant?
Yes. We connect to your calendar, telephony, and CRM through controlled, allow-listed tools, and irreversible actions require confirmation (a human-gate). ASR and synthesis run locally (faster-whisper + router), recordings are not kept by default (zero-retention), and the caller is told they are speaking with an AI assistant: GDPR and AI Act compliance designed in from the start.
Where do we start, and how good is its Polish speech recognition?
We start with one line and the 3–4 most common paths (status, hours, booking, escalation), using real recordings from your hotline. Let us be honest: ASR for Polish is good but not perfect: names, addresses, and numbers can be misheard, so the agent reads key details back aloud and, on any uncertainty or emotion, switches to a human immediately. That is a designed boundary, not an embarrassing exception.
voice-concierge · Docker / Kubernetes (On-Premises, Air-Gapped, or private EU VPC)
Executive Brief 1-Pager (Board & Procurement Summary)
The phone rings when there is no one to answer, and half the calls are the same questions about hours, status, and booking. Missed calls are lost appointments, and staffing a desk for routine work is expensive. Off-the-shelf voice bots lose context and never know when to hand a call to a human.
deploy: self-hostrecordings: zero-retentionresponse time: <2scall paths: 4 tryby
Target integrations:routerBGE-M3QdrantOllamafaster-whispern8n
→Related