# Answer a business's inbound calls, take details and book or route

> Answer a business's inbound calls, take details and book or route: 4 companies fully do it, 17 with a limit, 28 carry the tag. Every listing in Conversational voice agents with the reason and a source.

- Canonical: https://www.anchorterminal.com/niches/voice-agents.inbound-receptionist/
- Markdown: https://www.anchorterminal.com/niches/voice-agents.inbound-receptionist/index.md (~2,450 tokens)
- Slim: https://www.anchorterminal.com/niches/voice-agents.inbound-receptionist/index.min.md (~480 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/niches/voice-agents.inbound-receptionist/index.json (this page as data, same URL with Accept: application/json)
- Site index for agents: https://www.anchorterminal.com/llms.txt (the key docs and every listing in one file, at most 16 MB, about 4 million tokens: https://www.anchorterminal.com/llms-full.txt)
- API: https://www.anchorterminal.com/api/v1/index.json
- MCP: https://www.anchorterminal.com/mcp (the whole catalogue as tools, no key)
- DNS: dig +short TXT <slug>.grade.anchorterminal.com (the grade as a DNS TXT record, for a sandbox with no HTTP where names still resolve; also top., best.<category>. and <a>.vs.<b>. under the zone, and the same under grade.letme.dev)
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)
- Updated: 2026-11-15

Four or more companies' docs show a full fit.

| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |
| --- | --- | --- | --- | --- |
| 4 | 17 | 28 | 20 | 7 |

## What counts as done

The caller's name and number are captured and read back, a booking or lead row exists in the connected system when the agent says it does, and no booking is confirmed that the tool did not return.

Measured, for our own tests: the call is answered within two rings, and the digits are read back correctly.

How agents ask for it: answer my phone after hours; set up an AI receptionist that books appointments; take calls for my clinic and look the patient up

## Does the job

The docs show it meets the test above, and nothing we read says otherwise.

- **Bland AI API + MCP (Intelliga Corporation)**: Bland supplies numbers and SIP trunks; custom API tools take a 1 to 60 second timeout, speak a configured line while running, and return ok and message so a failed booking is visible to the agent. Transfers to a human are parameters.
- **LiveKit Agents (LiveKit Incorporated)**: LiveKit Phone Numbers or any SIP trunk route calls into a room; tools keep running and record results when speech interrupts, disallow_interruptions guards writes, and GetPhoneNumberTask reads numbers back by default on audio.
- **Synthflow API + MCP (AgentFlow AI GmbH)**: Numbers can be bought or brought over SIP, ring duration is configurable, business hours and real-time booking actions exist, and custom actions speak a failure message when the request fails. Sold only as an annual enterprise contract.
- **Vapi API + MCP (Vapi Inc.)**: Free US number or imported Twilio, Telnyx and SIP numbers; tools with request-start and delayed messages, 1 to 300 s timeout and a request-failed message. A 90-day plumbing receptionist with capture_lead and route_emergency tools reports 98 per cent of calls answered.

## Does the job, with a limit

It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.

- **Amazon Nova 2 Sonic (Amazon Web Services, Inc.)**: No phone numbers: calls arrive through a Twilio, Vonage or SIP bridge from an AWS sample. Async tools are never cancelled by caller speech, but digit readback and the final-utterance transcript on hang-up are not documented.
- **AssemblyAI Voice Agent API (AssemblyAI, Inc.)**: Inbound SIP from a Twilio number the customer already owns; no numbers sold. HTTP tools have per-tool timeout_seconds and errors reach the model as text; tool parameter patterns make the agent wait for a full phone number. Readback and hang-up transcript untested.
- **Bolna API + MCP (Voxlabs Private Limited)**: Numbers can be bought through the API or brought via Twilio, Plivo, Exotel, Vobiz or SIP, and API tools speak a pre_call_message. Two accounts show calls dropped at 10 to 12 seconds of silence and an agent stuck marked as speaking.
- **Deepgram Voice Agent API (Deepgram, Inc.)**: No numbers or SIP: calls come through a Twilio, Plivo, Genesys, Amazon Connect or AudioCodes bridge the developer runs. Barge-in cancels client function calls by default; irreversible tools need defer_until_eot.
- **ElevenLabs Agents API + MCP (Eleven Labs Inc.)**: Native Twilio and SIP trunking bring the business's numbers in. Webhook tools have a 5 second timeout floor and tool_error_handling_mode auto hides errors from non-native tools, so an account returns 200 with a spoken fallback to avoid a false confirmation.
- **fone.me (Fone)**: llms.txt says the agent answers calls on a Premium number and takes messages and booking requests from what the owner taught it, but requests land in the owner's inbox, not a connected booking system. Only the vendor's claim; guide pages need JavaScript.
- **Fresha AI Concierge (Fresha.com SV Ltd)**: Fresha says it picks up in under two seconds on an existing or new number, books, reschedules and cancels in the Fresha calendar, transfers to a set number and records and transcribes every call. Vendor marketing only, and only for Fresha Business accounts.
- **Gemini Live API (Google LLC)**: Function calling and transcription work, but Live has no numbers, SIP or carrier example, so a phone line must be bridged by the developer. Interrupted bookings can be re-issued (acc-14) and bridged calls drop at about 10 minutes without resumption (acc-15).
- **GPT-Live (OpenAI OpCo, LLC)**: Inbound SIP to sip.api.openai.com with delegated tools and spoken verified results, but the number must be bought from a SIP trunk provider such as Twilio and pointed at OpenAI first. Digit readback is undocumented.
- **Harmony.ai (monday.com Ltd.)**: Calls carry direction, outcome, collected fields and a transcript through the API and MCP, but the agent, its tools and booking are configured in the web app with no public reference; booking and number provisioning rest on the vendor's pages.
- **Inworld Realtime API (Theai, Inc. dba Inworld AI)**: A Twilio Media Streams guide answers calls with tools and fillers, but the developer buys the Twilio number and runs the bridge server; Inworld supplies no numbers or SIP. Digit readback is undocumented.
- **OpenAI Realtime API (OpenAI OpCo, LLC)**: Inbound SIP calls are accepted via webhook, but the developer brings the trunk and number; digits are swapped or dropped on read-back (closed as a model limit) and the final utterance can be lost when the caller hangs up.
- **Pipecat (Daily)**: Daily phone numbers, Twilio, Telnyx and Plivo serializers take inbound calls, but tools default to cancel_on_interruption=True; a clinic agent told a caller a listed patient was not on file until every tool was set to False.
- **Retell AI API + MCP (Retell AI Inc.)**: Numbers are sold, custom functions run to completion when interrupted, with speak-during-execution and a 1 to 600 s timeout; but an interrupted function's result is not read back at that moment, and an agency monitor reports calls that sound fine while no booking is made.
- **Rime Prism (Rime Labs, Inc.)**: Tools and fillers work, but Prism has no numbers or SIP and takes only 16 kHz PCM16, so the developer bridges a carrier and converts mu-law both ways before a call can be answered.
- **Ultravox Realtime API (Theai, Inc. dba Inworld AI)**: Tools run server-side and the agent speaks again after the result, but Ultravox sells no numbers: inbound calls arrive through the developer's Twilio, Telnyx, Plivo, Exotel or SIP account, and the default tool timeout is 2.5 s.
- **Vaani Developer Platform (Vaani AI Research Private Limited (Bengaluru, India, CIN U62099KA2025PTC201212))**: Numbers from Twilio, Plivo or Exotel or a BYO SIP trunk, Cal.com booking and cURL-defined API tools exist, but the docs give no tool timeout, error handling or cancellation behaviour, so a wrong spoken outcome cannot be ruled out.
- **Vocily AI (VOCILY AI PRIVATE LIMITED)**: Numbers are rented from Vocily or imported from Plivo or VoBiz, and custom API and Cal.com tools have an editable pre-action line and a failure line. Indian rules require a registered business behind every number, and tool timeouts are not stated.
- **Vogent API (Monoid, Inc.)**: Numbers can be purchased or imported over SIP through the API, and functions have started and waiting messages, but no tool timeout, error message or cancellation behaviour is documented, and API functions are POST only.

## Couldn't check

The docs were closed to us, or silent. These carry no mark either way.

- **Hume EVI (Empathic Voice Interface) (Hume AI, Inc.)**: EVI docs withdrawn: every EVI page, including the overview and tool-use guide, redirects to the Audio Decisions overview, and the EVI AsyncAPI file returns 404. The last assessment recorded EVI access ending 13 November 2026.

- not this job: CueLight (sshah03 (individual GitHub account, no name or organisation stated)): CueLight is an end-of-turn detector only: no call handling, telephony, tools or speech output.
- not this job: Pipecat PhoneLLM (Daily): A text model for the LLM step only; answering calls, numbers and booking rows come from the framework and telephony around it.
- not this job: PlaceCall (VOYGR Tech, Inc.): Outbound only: POST /calls dials a US business for the user; there is no inbound number or answering agent.

## What stops agents finishing

- no phone numbers or SIP on the platform, so a carrier must be bridged first
- a tool call cancelled by caller speech makes the agent state a wrong outcome
- webhook tool timeout floor above the latency the business will accept

## Around it

Before: telephony.inbound-number, crm.contact-lookup. After: booking.reservations, crm.create-lead, messaging.sms-confirm.

## Where this comes from

Weight 20 of 100 in Conversational voice agents: 7 of 31 accounts (acc-05, acc-09, acc-10, acc-18, acc-19, acc-29, acc-30) plus every vendor use-case page read names it first

As data: https://www.anchorterminal.com/api/v1/use-cases/voice-agents.inbound-receptionist.json, or the MCP tool get_use_case with id voice-agents.inbound-receptionist. Draft, read 2026-11-12 from public docs and first-hand accounts, not yet reviewed by a person.

---

Query this catalogue directly over MCP: https://www.anchorterminal.com/mcp (Streamable HTTP, no key). How: https://www.anchorterminal.com/agents/#mcp
