Silux Voice vs the field
Self-hostable, multi-engine AI receptionist: BYO-LLM + multi-provider routing, local-only LLM mode, built-in voice cloning, 6 TTS + 3 STT engines, PAYG $0.04/min and a real Free Forever tier — strongest on price, flexibility and data sovereignty. Uniquely, it remembers the callers it has already spoken to — with confidence scoring, provenance and a per-tenant off switch — so customers stop repeating themselves.
82 capabilities across 33 AI-receptionist and voice-AI platforms. Silux Voice is the only platform that wins every row. Showing our latest published dataset.
Why Silux Voice wins
The capabilities no competitor in this table matches — cross-call caller memory, live in-call booking into your preferred calendar, built-in voice cloning, and a Free Forever tier rather than a 7-day trial.
- ✓Cross-call caller memory — returning callers never repeat themselves
- ✓Confidence-scored, provenance-tracked memory that confirms instead of asserting
- ✓Fabrication guard verifies every reply before it is spoken
- ✓One-switch memory off, caller opt-out and spoken “forget my details” erasure
- ✓Self-host / on-prem / data-sovereignty deployment
- ✓BYO-LLM per tenant + multi-provider LLM routing with fallover
- ✓Local-only LLM mode (no cloud hop)
- ✓Built-in voice cloning (XTTS-v2)
- ✓6 TTS + 3 STT engines with automatic fallback
- ✓$0.04/min PAYG + Free Forever tier + try-it-now sandbox
- ✓Per-country currency + international pricing
- ✓PCI DTMF payment-by-phone + redaction, per-tenant cost ledger
The price you see is the price you pay
Every other AI receptionist bills per minute or per call. Silux Voice bills per booked job — the way you price your own work. Jobs booked, not minutes talked. See our pricing →
Per minute / per call
The meter runs on hold music, spam and wrong numbers — you pay whether or not a real job is ever booked.
0.05 – 0.06 / min
A small monthly base fee, then you pay when a job is actually booked — predictable, fair, and priced the way you price your own work.
Others start every call from scratch. We pick up where you left off.
A handful of platforms can read a name out of a CRM record you typed in yourself. None of them build memory from the conversation, score how sure they are, or let you switch the whole thing off in one click. See how caller memory works →
"Can I take your name?"
- · Memory only if somebody typed it into a CRM first
- · No sense of how reliable a detail is
- · No audit trail when something is corrected
- · No caller-side opt-out on the call itself
"Hello Jane, welcome back."
- · Learned from the call itself — no admin afterwards
- · Every fact scored, sourced and expiring on its own
- · Unsure details are confirmed, never asserted
- · Off in one switch, and callers can opt out mid-call
The comparison matrix
11 columns × 82 rows. Showing the top 10 competitors by capability coverage; see all 33 below. Scroll sideways to compare — the capability and Silux Voice columns stay pinned.
| Capability | Silux Voice | Smith.aiUS · GLOBAL | CloudTalk (CeTe)EU · GLOBAL | DialzaraUS · GB | Rosie AIUS · GB | RingCentral AIRUS · GB · GLOBAL | ServiceAgentUS | NextPhoneUS | MoneypennyGB · US | Ruby ReceptionistsUS | Retell AIUS · GLOBAL |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Core answering | |||||||||||
| 24/7 AI call answeringRuby/AnswerConnect/Nexa are human-led; AI is tooling, not the answerer. | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | partial | ✓ |
| Natural, human-like voice | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | ✓ |
| Unlimited concurrent calls | ✓ | — | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | — | ✓ |
| Appointment booking / calendar sync | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — |
| Message-taking with custom questions | ✓ | ✓ | — | ✓ | ✓ | ✓ | — | ✓ | ✓ | ✓ | — |
| Lead qualification / screening | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | — | ✓ | — |
| Smart call routing / transfer | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — |
| Spam / robocall filtering | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | ✓ | — | — | — |
| Call summaries + transcripts | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | ✓ | ✓ | ✓ | ✓ |
| Emergency / urgency keyword routing | ✓ | — | — | — | — | — | ✓ | ✓ | — | — | — |
| Channels | |||||||||||
| Bilingual / multi-language support | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | ✓ | ✓ | partial | — |
| Live in-call translation (both directions)Smith.ai does EN<->ES real-time chat translation; none offer in-call voice translation as a bundled add-on. | ✓ | partial | — | — | — | — | — | — | — | — | — |
| Outbound calling / campaigns | ✓ | ✓ | ✓ | ✓ | — | — | — | — | ✓ | — | ✓ |
| Two-way SMS agent | ✓ | — | ✓ | ✓ | ✓ | — | ✓ | — | — | — | — |
| WhatsApp channel | ✓ | — | ✓ | — | — | — | ✓ | — | — | — | — |
| Website chat / chatbot widget | ✓ | ✓ | — | ✓ | ✓ | — | ✓ | — | — | ✓ | — |
| Native mobile app | ✓ | — | — | — | ✓ | — | — | ✓ | ✓ | ✓ | — |
| Integrations | |||||||||||
| Native CRM integrations | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | — | — |
| Zapier / Make / n8n integration | ✓ | ✓ | — | ✓ | ✓ | — | — | ✓ | — | — | — |
| Field-service (FSM) integrationsJobber / Housecall Pro / ServiceTitan; Silux ships 4 named FSM integrations. | ✓ | — | — | — | ✓ | — | ✓ | — | — | — | — |
| Open API / SDK | ✓ | ✓ | ✓ | — | — | — | — | — | — | — | ✓ |
| Outbound webhooks (event push) | ✓ | ✓ | ✓ | ✓ | — | — | — | — | — | — | ✓ |
| Knowledge-base auto-ingest from website | ✓ | — | — | — | ✓ | — | ✓ | — | — | — | — |
| Payment collection over the phoneSilux adds PCI DTMF payment-by-phone + Stripe agent flow. | ✓ | ✓ | — | — | — | — | ✓ | — | — | ✓ | — |
| LLM & engine flexibility | |||||||||||
| Bring-your-own LLM endpoint per tenantRetell allows BYO LLM at build time; only Silux offers per-tenant BYO-LLM routing as a product feature. | ✓ | — | partial | — | — | — | — | — | — | — | ✓ |
| Multi-provider LLM routing + falloverSilux routes across DeepSeek / Gemini / OpenAI / Grok / local with priority + fallback. | ✓ | — | partial | — | — | — | — | — | — | — | — |
| Local-only LLM mode (no cloud hop)No competitor offers a true local-only per-tenant SLA flag. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Self-host on your own hardwareEvery competitor is cloud-only SaaS; Silux can be deployed on-prem. | ✓ | — | — | — | — | — | — | — | — | — | — |
| On-prem / data-sovereignty deploymentSilux-only: run entirely inside the customer's network. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Multi-engine TTS with automatic fallbackSilux runs 6 TTS engines (Coqui/Kokoro/Piper/Mimic3/Cartesia/ElevenLabs). | ✓ | — | partial | — | — | — | — | — | — | — | — |
| Multi-engine STTSilux runs 3 STT engines (Whisper/Vosk/Deepgram Nova-2). | ✓ | — | partial | — | — | — | — | — | — | — | — |
| Built-in voice cloning (XTTS-v2)Vellum answers in your cloned voice but does not expose cloning as a slot; Silux includes a clone slot. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Sub-100ms TTS first-token latency optionSilux Pro+ via Cartesia Sonic (~90ms); competitors sit at 200-600ms. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Pricing & access | |||||||||||
| Free Forever tier (real value, not a trial)Competitors offer 7-14 day trials at best; Silux has a 30 min/mo free-forever tier. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Pay-as-you-go (no monthly minimum)Silux PAYG is $0.04/min, cheapest in category. | ✓ | — | — | ✓ | ✓ | — | ✓ | — | — | — | ✓ |
| Try-it-now public sandbox (no signup)Silux /try lets you speak to it with no signup; competitors gate behind demo/booking. | ✓ | — | — | — | — | — | — | — | — | — | — |
| International auto-converted pricing (multi-currency)Silux resolves per-country currency by cookie/geo across 12+ currencies. | ✓ | — | partial | — | — | partial | — | — | — | — | — |
| White-label / reseller program | ✓ | — | — | — | — | — | — | — | ✓ | — | — |
| Compliance & operations | |||||||||||
| BAA available | ✓ | — | — | — | — | — | — | — | — | — | — |
| Call recording + searchable transcription | ✓ | ✓ | ✓ | — | — | ✓ | — | — | — | ✓ | — |
| PCI DTMF redaction on recordingsSilux-only defence-in-depth toggle. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Encrypted storage + audit trails | ✓ | ✓ | ✓ | ✓ | — | ✓ | — | — | — | — | — |
| Per-tenant LLM cost ledgerSilux-only usage/cost accounting per tenant. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Caller memory | |||||||||||
| Recognises a returning caller by their numberMost recognition in the category is a CRM/record lookup rather than learned memory; human-led services rely on their agents' notes. | ✓ | ✓ | partial | — | — | — | — | — | partial | partial | — |
| Builds memory from the conversation itself — no manual CRM entrySilux Voice extracts facts from what the caller actually said, during the call. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Returning callers never re-answer questions already on fileKnown slots are checked before the AI is allowed to ask for them. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Confidence score on every remembered fact — confirms instead of assertingBelow 0.5 a fact never reaches the model; 0.5-0.75 is spoken as a question. Not advertised on their public product pages as of Aug 2026. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Provenance + source-trust ranking on every factA guess can never overwrite something the caller stated aloud. Not advertised on their public product pages as of Aug 2026. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Corrections supersede with a full audit trailOld values are retained and linked, not overwritten. Not advertised on their public product pages as of Aug 2026. | ✓ | — | — | — | — | — | — | — | — | — | — |
| The LLM can never write to memory during a callExtraction is deterministic, so an invented detail cannot become a remembered fact. Not advertised on their public product pages as of Aug 2026. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Automatic fact expiry + tenant retention sweepReported issues expire at 120 days, promises at 45. Not advertised on their public product pages as of Aug 2026. | ✓ | — | — | — | — | — | — | — | — | — | — |
| One-switch memory off, with no shadow profile keptOff means nothing recalled and nothing written. Not advertised on their public product pages as of Aug 2026. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Caller-side opt-out and spoken “forget my details” erasureErasure happens on the call, not via a support ticket. Not advertised on their public product pages as of Aug 2026. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Per-tenant caller isolation enforced in every queryMemory is keyed to tenant + number together, never the number alone. Not advertised on their public product pages as of Aug 2026. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Replies verified against grounded facts before they are spokenPrices, times, days and durations must trace back to a real source or the reply is replaced. Not advertised on their public product pages as of Aug 2026. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Executive Assistant | |||||||||||
| Keypad menu into a department-specific AIPress 1 for sales, 2 for accounts — and that department's own AI answers, reading only that department's knowledge. Most products route to a human or run one general AI. | ✓ | — | — | — | — | partial | — | — | — | — | — |
| Separate knowledge and rules per departmentAccounts reads accounts information only. Narrow scope is the accuracy mechanism — a billing question that can reach sales content comes back subtly wrong. | ✓ | — | — | — | — | — | — | — | — | — | ✓ |
| Tries several extensions per department, in orderA department is rarely one phone. Ringing only the first number is how 'we transfer your calls' quietly means 'we ring a desk nobody sits at'. | ✓ | partial | — | — | — | partial | — | — | — | — | — |
| Takes the call back when nobody answersRather than dropping the caller into voicemail, the AI returns, says honestly that nobody is picking up, and takes their details. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Alerts graded by urgency (text/WhatsApp vs email)Urgent goes to a phone, routine goes to an inbox. A business texted about everything stops reading the texts. | ✓ | partial | — | — | — | — | — | — | — | — | — |
| Never tells the caller a ticket was openedThe caller is told a person will call them back. The bookkeeping happens behind them. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Quality | |||||||||||
| Test suites for your own receptionistWrite down what callers actually say and what should happen, then run it whenever anything changes. | ✓ | ✓ | — | — | — | — | — | — | — | — | partial |
| Tests generated from your own knowledgeChange a price and the tests for it are written and run automatically. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Regression alerts when results get worseTold when today's results are worse than yesterday's, and what changed in between. | ✓ | partial | — | — | — | — | — | — | — | — | — |
| Improvement | |||||||||||
| Repeated questions grouped into a suggested answer'18 callers asked about Saturday opening' — with the answer drafted for approval. | ✓ | partial | — | — | — | — | — | — | — | — | — |
| Receptionist quality score out of 100Six sub-scores from six independent sources, so a weak area is visible rather than averaged away. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Reviews its own performance and proposes the fixWorks out what is driving transfers, suggests the change, and shows the before and after. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Trust & safety | |||||||||||
| Approval required before risky actionsEvery action sorted into low, medium and high risk. Refunds and cancellations wait for a person; booking an appointment does not. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Warns a human before the caller asks for oneSpots a conversation heading for escalation and briefs a person early. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Warm handover carrying what was already promisedWho is calling, why, what has been said, and what the AI already committed to — so the human does not contradict it. | ✓ | ✓ | — | — | — | — | — | — | partial | partial | — |
| Time-to-human measured and publishedSeconds between asking for a person and a person speaking. Nobody else reports it. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Reporting | |||||||||||
| Revenue traced from call to completed job'£18,400 of work' rather than '300 calls answered'. | ✓ | partial | — | — | — | — | — | — | — | — | — |
| Business insights, not just call countsWhat people are ringing about and what is changing week to week. | ✓ | — | — | — | — | partial | — | — | — | — | — |
| Sales opportunities spotted inside callsAn ageing boiler, an expiring contract, a second property — flagged from what the caller said. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Competitor mentions tracked and aggregatedWho you are compared with, on what, and whether you won. | ✓ | — | — | — | — | — | — | — | — | — | — |
| One connected history per customerCall, text, appointment, enquiry and follow-up in one timeline rather than five lists. | ✓ | partial | — | — | — | — | — | — | — | — | — |
| Screening | |||||||||||
| Nuisance-caller reputation shared across every businessA number thirty businesses reported is blocked before it reaches the thirty-first. | ✓ | partial | — | — | — | — | — | — | — | — | — |
| Omnichannel | |||||||||||
| Website chat and phone call joined into one conversationSomeone chats on your site then rings — the receptionist already knows what about. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Multilingual | |||||||||||
| Route by language to someone who speaks itAnd it will only promise a call back in a language you have said somebody speaks. | ✓ | partial | — | — | — | — | — | — | — | — | — |
| Commercial | |||||||||||
| Buy one feature at a time, or a packNo forced upgrade to a whole tier for one capability. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Turn any feature off yourselfYour switch beats your plan. If you do not want the AI doing something, it does not. | ✓ | — | — | — | — | — | — | — | — | — | — |
| Roles and per-area permissions for your teamFive roles across sixteen areas — calls, recordings, customers, calendar, billing, and the rest. | ✓ | ✓ | — | — | — | ✓ | — | — | — | — | — |
Legend: ✓ supported · partial partially/conditionally · — not offered. Competitor values reflect published product information; Silux Voice ships every capability in this table.
Every platform we compare against
All 33 AI-receptionist and voice-AI products in the dataset, with their positioning and regions.
Smith.ai
US · GLOBALHybrid AI + 24/7 North-America human receptionists; deep integrations, payment collection, HIPAA.
RingCentral AIR
US · GB · GLOBALAI receptionist built into (or standalone on) a full cloud phone system; sub-700ms carrier-network latency.
Voksha
USFocused low-cost pure-AI SMB/SaaS answering tool.
Goodcall
USNo-code drag-and-drop call-flow builder; unlimited minutes billed by unique callers.
Dialzara
US · GBMulti-channel pure-AI: voice + outbound + two-way SMS + web chatbot on one 'brain'; strong compliance.
Rosie AI
US · GBPure-AI with unlimited concurrent calls, auto-training, native iOS app; sub-$0.20/min effective.
My AI Front Desk
USPure-AI with white-label reseller program; fast non-technical setup.
SkipCalls
USFlat annual pricing, unlimited minutes, no overages; 60-second setup.
Slang.ai
USPurpose-built for restaurants/hospitality; deep reservation-platform integrations.
AgentZap
USAppointment-driven pure-AI across 90 industries; booking-first workflow.
NextPhone
USFlat-rate unlimited inbound, sub-5s pickup, native mobile apps.
Vellum
USPersonal AI assistant answering in your own cloned voice with persistent memory across surfaces.
Ruby Receptionists
USHuman-led with AI-enhanced tooling; transparent inclusive pricing, best-in-class app.
AnswerConnect
US · GBHuman-first (no bots) 24/7 receptionists with e-commerce focus.
Nexa
USHuman-led receptionists trained on home-service/contractor call types.
Upfirst
USSimple generalist pure-AI answering on the first ring; young company.
Bookipi AI Receptionist
GLOBALPure-AI front desk inside Bookipi's SMB/invoicing ecosystem; 3-step setup.
Weave / TrueLark
USAI receptionist inside Weave's unified dental/medical comms platform; HIPAA + BAA.
ServiceAgent
USHome-services pure-AI (on Twilio + Retell) with Stripe payments and live takeover.
Numa
USPurpose-built for auto dealerships; understands automotive terminology and DMS.
Dialpad AI
US · GLOBALAI receptionist bundled inside an all-in-one UCaaS platform.
Retell AI
US · GLOBALDeveloper BYOK voice platform — a toolkit to build agents, not a finished product.
Sully.ai
USModular AI medical employees; deep EHR write-back and clinical decision support.
Talkie.ai
US · EUMedical voicebots pre-trained on healthcare transactions; deep EHR connectivity.
DoctorConnect (ARIA)
USAI medical receptionist inside a patient-engagement platform; 150+ EHR/PMS integrations.
CloudTalk (CeTe)
EU · GLOBALAI voice agent inside a VoIP platform; 60+ languages, LLM/voice-engine choice, deep conversation intelligence.
IONOS AI Receptionist
EU · GBLowest EU/UK entry pure-AI (from £9/mo), built on Newo.ai, for lower call volumes.
Moneypenny
GB · USHybrid proprietary AI + UK/US human receptionists; solutions-architect onboarding.
Marlie
GBMost self-configurable UK AI-first option with optional human escalation.
BookedSolid
GBPure-AI for UK private healthcare clinics; deep PMS booking integration.
ARROW
GBManaged AI receptionist for UK trades; flat-rate, managed setup.
Lotusbrains Studio
GBBespoke custom-built AI voice-agent receptionists (4+ years); deep per-client tailoring.
AIAnswerPhone / Hand On Web
GBUK pure-AI with British accents, WhatsApp notifications, GDPR UK infrastructure.
Win every row for your business
14-day free trial, $0.05/min pay-as-you-go. No credit card to start.