Phonely
by Phonely Ltd.
Voice agents running on Alma, a model trained on ten million real phone calls
Phonely is a voice-AI platform for building and running production phone agents across calls, SMS, chat and API, aimed at contact centres and high-volume inbound operations. It runs on Alma, the company's own voice-tuned language model trained on more than ten million real phone conversations, which it also sells to teams building voice agents elsewhere.
Phonely is a voice-AI platform for building production phone agents that handle inbound and outbound calls, plus chat, SMS and API-driven conversations, from a no-code agent builder. Founded in 2023 and a Y Combinator Summer 2024 company, it is unusual in the category for running its own model rather than renting a general-purpose one. On 1 September 2026 it launched Alma, a large language model trained on more than ten million real phone conversations and built for the actual failure modes of telephony — interruptions, background voices, crosstalk and transcription errors — rather than clean turn-taking. Phonely reports Alma reaching first token in under 185 milliseconds against roughly 500 milliseconds for GPT-4.1, and a blended cost of $0.55 per million tokens against $3.50 for GPT-4.1 and $5.63 for GPT-5.4. Alma now serves all of Phonely's own agent traffic, handling millions of calls a month, and is available to teams building voice agents elsewhere, working alongside their existing transcription and text-to-speech providers. The wider platform adds call recording, transcription and AI analytics, more than 1,000 voices with cloning and over 100 languages with regional accents, and roughly 22 named integrations including HubSpot, Salesforce, Google Calendar, Mailchimp, Airtable, Zapier, REST and webhooks. The Enterprise tier adds SIP trunking, custom fine-tuning, dedicated account management and a HIPAA BAA. Inference runs on fine-tuned models served via Groq LPU hardware. Phonely raised a $16 million Series A led by Base10 Partners in April 2026, roughly $19 million in total, and names Etech Global Services, TSA Group, Engage CX, Lifelike Health and Signpost among its customers.
The head of contact centre or customer operations running high, predictable inbound call volume, where turn-taking latency and per-minute cost are the two things that decide whether voice AI is viable.
Sub-second, voice-native agents at a published per-minute price, on a model tuned for messy real calls rather than clean transcripts.
At a Glance
- Category
- Audio & Voice
- Pricing
- Freemium, Subscription, Usage-based
- Target Market
- Contact Centre Leaders, VPs of Customer Operations, CTOs, Enterprise Developers, BPO Operators
- Deployment
- Cloud-only, API-based
- Founded
- 2023
- Headquarters
- San Francisco, United States
- Customers
- 10,000+ businesses and 120,000+ calls per day, both vendor-reported
Key Features
- ✓Alma voice-native LLM
A model trained on 10M+ real calls that handles interruptions, crosstalk and transcription errors rather than assuming clean turn-taking.
- ✓Sub-185ms time to first token
Vendor-measured latency roughly a third of GPT-4.1's, which is what removes the unnatural pause a caller hears.
- ✓No-code agent builder
A visual editor for conversation flows, so operations staff can ship and revise an agent without pulling in developers.
- ✓1,000+ voices with cloning and 100+ languages
A large voice library with cloning and regional accents, letting one platform serve several markets and brand voices.
- ✓Unified multi-channel flows
The same agent logic drives phone, SMS, chat and API conversations instead of being rebuilt separately per channel.
- ✓Enterprise telephony and fine-tuning
SIP trunking, custom model fine-tuning, dedicated account management and a HIPAA BAA are gated to the Enterprise tier.
- ✓Published pricing with a free tier
Four public tiers with a full feature matrix, which is rare in voice AI and makes budgeting possible before a sales call.
Capabilities
Use Cases
- •Inbound call deflection
Answering FAQs, checking order status and routing callers so human agents only take the calls that genuinely need them.
- •Appointment scheduling
Agents book, reschedule and confirm appointments straight into Google Calendar or a connected CRM during the call.
- •Outbound lead qualification
Calling inbound leads within seconds, qualifying against criteria and handing warm ones to sales with full call context.
- •BPO and contact-centre scale-out
Etech Global Services and Engage CX run tens of thousands of calls a day through Phonely agents.
- •Regulated intake calls
Healthcare and insurance intake handled under an Enterprise HIPAA BAA, with transcripts and analytics retained for quality review.
Ideal For
Best For
- ✓Contact centres and BPOs running high, predictable inbound call volume
- ✓Teams that want a voice-tuned model rather than stitching a general-purpose LLM onto a telephony stack
- ✓Latency-sensitive deployments where the unnatural pause before an agent replies is the main quality complaint
- ✓Builders who want to use Alma with their own existing speech-to-text and text-to-speech providers
- ✓Operations teams that need published pricing to budget before entering a sales cycle
Not Ideal For
- ✗Buyers who require third-party-audited benchmarks — every headline latency, cost and accuracy figure is self-reported with no published methodology
- ✗Regulated organisations that need to inspect SOC 2 or HIPAA attestations before signing, which Phonely does not make publicly available
- ✗Variable-volume SMBs, where $0.25-$0.35/min overage on the small tiers makes a spiky month expensive fast
- ✗Companies that need a full business phone system for human agents, not only an AI agent layer
- ✗Buyers who rely on an independent review base, since G2 and Capterra profiles exist with effectively no usable ratings
Integrations
Deployment
Market & Ratings
10,000+ businesses and 120,000+ calls per day, both vendor-reported
Market Analysis
Pros
- ✓Owns its model, so latency and per-token cost are not hostage to a frontier vendor's roadmap or price list
- ✓Pricing published across four tiers with a full feature matrix, unusual among AI voice startups
- ✓Genuinely free tier of 100 minutes a month, with unlimited concurrent calls on every plan
- ✓Named enterprise references operating at real volume, reported at 15,000 to 60,000 calls a day
- ✓Alma is usable alongside existing transcription and TTS vendors rather than forcing a whole-stack replacement
Cons
- ✗Every headline benchmark — latency, cost and accuracy — is self-reported with no third-party audit or published methodology
- ✗Accuracy is quoted inconsistently across the vendor's own pages (99%, 99.2% and 99.7%) and call-detectability claims range from 70% to 90%
- ✗The advertised 'as low as 5c/min' applies only to Enterprise; self-serve tiers work out around $0.20/min on committed minutes, roughly four times that
- ✗SOC 2 and HIPAA are claimed but the attestations are not publicly inspectable, and no data residency or call-retention policy is published
- ✗Effectively no independent review base: G2 and Capterra profiles carry no usable ratings, and Hacker News has a single low-engagement submission about the Alma launch
- ✗Integration documentation does not distinguish native connectors from Zapier-brokered ones, and enterprise minimums and SLA terms are unpublished
- ✗Independent reviewers report setup complexity for production deployments exceeding the marketing claim that most businesses start in under five minutes
Pricing
Free
$0
- ✓100 minutes/month (~50 calls)
- ✓100+ voices and languages
- ✓AI agent builder
- ✓Email and SMS notifications
Starter
From $33/mo
- ✓250 minutes/month
- ✓$0.25/min overage
- ✓Premium voices
- ✓Transcripts and AI analytics
- ✓Spam filtering
Professional
From $100/mo
- ✓750 minutes/month
- ✓$0.30/min overage
- ✓Conversational AI SMS
- ✓Advanced API integrations
- ✓CRM integrations
Enterprise
Contact for pricing
- ✓From $0.05/min at volume
- ✓Full agent buildout
- ✓Custom fine-tuning
- ✓SIP trunking
- ✓HIPAA BAA
- ✓Slack support
Pricing is published in full, which is rare in this category. Free gives 100 minutes monthly; Starter is $50/month ($33 billed annually) for 250 minutes with $0.25/min overage; Professional is $150/month ($100 annually) for 750 minutes with $0.30/min overage; Enterprise is quoted from $0.05/min at volume. Note that the self-serve tiers work out near $0.20/min on committed minutes, roughly four times the advertised enterprise rate, and enterprise minimums, contract length and SLA remedies are not published.
Security & Compliance
Sources
This page was written from 5 sources, 3 on domains other than phonely.ai.
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
Smallest.ai
Sub-second voice AI — small, specialised speech models for real-time enterprise voice agents
Ringg AI
Multilingual voice, chat and WhatsApp agents for high-volume enterprise conversations
Rime
Text-to-speech trained on studio-recorded conversation, built for regulated call volume
Vapi
Voice AI infrastructure for developers who want control, not a packaged agent