Category · Voice & speech
Conversational voice-agent APIs
Platforms that run a spoken conversation for you, either with one speech-to-speech model or a configurable pipeline of speech-to-text, a language model and text-to-speech. Compared on response latency, handling interruptions, tool calls, recovery after a misunderstanding and cost per completed call.
Best conversational voice-agent APIs, the shortlist583 head-to-head comparisons
Capability keys voice.agent · voice.speech-to-speech · voice.pipeline · voice.tools · voice.telephony · All tools
letme.dev/voice.agent picks the top-graded tool in this list and says how to call it direct (how letme works).
The same listing from the live API. Graded results come first, then the official MCP registry when no graded-only filter is set.
https://www.anchorterminal.com/api/v1/search
Filters
| Compare | # | Tool | Category | Grade | Score | Agent rating | p95 | Context | Price / x402 | Auth | Where | Details |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 93 | LiveKit AgentsLiveKit Incorporated · Agent framework | Voice agents | BB | 76.5 | 3.0 (4) | n/a | n/a | Free · OSS | API key | Library | ||
|
LiveKit Agents is LiveKit's open-source Python and Node.js framework for realtime voice and video agents that join LiveKit rooms, with STT, LLM, TTS and realtime-model plugins, turn detection, MCP tools and telephony through LiveKit SIP. Top strength Apache-2.0 Python and Node.js packages, with 18 Python releases in the 90 days to 1 November 2026 Top weakness The turn-detector models are under a proprietary licence that forbids use outside LiveKit Agents and training other models on their output |
||||||||||||
| 124 | ElevenLabs Agents API + MCPElevenLabs · HTTP API | Voice agents | BB | 74.9 | 3.5 (2) | n/a | n/a | $22 / mo | OAuth or key | Hosted | ||
|
ElevenAgents (formerly Conversational AI) runs hosted voice agents as a pipeline of a fine-tuned ElevenLabs ASR model, an LLM of your choice or your own, ElevenLabs TTS and a proprietary turn-taking model. Top strength API keys scoped to endpoint groups, with per-key credit limits and service accounts Top weakness $0.08 a minute excludes LLM tokens and carrier minutes |
||||||||||||
| 131 | Deepgram Voice Agent APIDeepgram · HTTP API | Voice agents | BB | 74.7 | 3.0 (2) | n/a | n/a | Pay per use | API key | Hosted and local | ||
|
One WebSocket that runs Deepgram STT (Flux or Nova-3), a managed or bring-your-own LLM and Deepgram or third-party TTS, with turn-taking, barge-in and function calling. Top strength $0.075 a minute all in on Standard, $0.050 with your own LLM and TTS, $200 of credit with no card Top weakness No phone numbers or SIP, so you bridge Twilio or another carrier yourself |
||||||||||||
| 170 | OpenAI Realtime APIOpenAI · HTTP API | Voice agents | BB | 73.6 | 3.3 (4) | n/a | n/a | Pay per use | API key | Hosted | ||
|
OpenAI's Realtime API runs spoken conversations with speech-to-speech models such as `gpt-realtime-2.1`. Clients connect over WebRTC, WebSocket or SIP, and sessions support interruptions, function calling and remote MCP tools. Top strength The OpenAPI 3.1 file in Top weakness Realtime was degraded in two API-wide incidents, for an hour on 17 September and for 5 hours 21 minutes on 29 September 2026. |
||||||||||||
| 241 | GPT-LiveOpenAI · HTTP API | Voice agents | BB | 71.9 | 3.5 (4) | n/a | n/a | Pay per use | API key | Hosted | ||
|
GPT-Live is OpenAI's full-duplex voice API (`gpt-live-1`), which listens while it speaks and delegates reasoning and tools to a Responses model, the Decisions API or the developer's own agent. Clients connect over WebRTC, WebSocket or SIP. Top strength The OpenAPI 3.1 file has seven Top weakness The Live component on status.openai.com has data only from 16 October 2026, so the five weeks after general availability have no readable history. |
||||||||||||
| 335 | Bland AI API + MCPBland AI · HTTP API | Voice agents | B | 69.5 | 3.0 (2) | n/a | n/a | $299 / mo | API key | Hosted and local | ||
|
Phone-first voice-agent platform that runs its own speech recognition, language model, TTS and telephony, billed as one per-minute rate. Top strength One per-minute rate covering STT, LLM, TTS and telephony, $0.14 on Start and $0.12 on Build Top weakness No OpenAPI document |
||||||||||||
| 364 | PipecatDaily · Agent framework | Voice agents | B | 68.9 | 3.0 (4) | n/a | n/a | Free · OSS | None | Library | ||
|
Pipecat is Daily's open-source Python framework for real-time voice and multimodal agents, built as pipelines of transport, speech-to-text, LLM and text-to-speech services, with client SDKs and optional hosting on Pipecat Cloud. Top strength BSD-2-Clause framework, with 23 speech-to-text, 25 LLM, 31 text-to-speech and 6 speech-to-speech services listed in the README Top weakness No tool-approval or confirmation step, and no prompt-injection guidance, for tools that act on a caller's speech |
||||||||||||
| 371 | Retell AI API + MCPRetell AI · HTTP API | Voice agents | B | 68.7 | 3.5 (2) | n/a | n/a | $2 / mo | API key | Hosted | ||
|
Hosted platform for phone and web voice agents, built as single-prompt agents or node-based conversation flows. Top strength Published per-minute price for every component, from $0.07 a minute all in Top weakness Call data kept indefinitely unless retention is set per agent |
||||||||||||
| 397 | Amazon Nova 2 SonicAmazon Web Services · HTTP API | Voice agents | B | 68.2 | none | n/a | n/a | Pay per use | API key | Local | ||
|
Amazon Nova 2 Sonic is AWS's speech-to-speech model for voice agents on Amazon Bedrock. Clients stream audio and text events both ways over the `InvokeModelWithBidirectionalStream` API, with turn-taking, interruptions, asynchronous tool calls and multilingual voices. Top strength No Bedrock event on the AWS Health history in the 90 days from 8 August 2026 Top weakness Connections end after 8 minutes, so long calls need the session-continuation pattern |
||||||||||||
| 468 | Vapi API + MCPVapi · HTTP API | Voice agents | B | 66.9 | 2.5 (2) | n/a | n/a | $29 / mo | API key | Hosted and local | ||
|
Developer platform for phone and web voice agents. Top strength Any mix of transcriber, model and voice provider, or your own keys and endpoints Top weakness Real per-minute cost depends on provider choices |
||||||||||||
| 654 | Pipecat PhoneLLMDaily · Model API | Voice agents | B | 62.9 | 2.8 (4) | n/a | n/a | from $0.06 / 1M in | API key | Hosted | ||
|
Pipecat PhoneLLM is Daily's open-weight text model for the language step of phone voice agents, with tool calling. Alpha 2, released 15 October 2026, comes as Mini (30B, on Workers AI) and Max (120B, weights and dedicated endpoints). Top strength Mini is on Workers AI at $0.06 per million input and $0.24 per million output tokens, inside the 10,000 free neurons a day Top weakness Labelled Alpha 2 by Daily, with no deprecation policy, support period or changelog of its own |
||||||||||||
| 674 | Inworld Realtime APIInworld AI · HTTP API | Voice agents | B | 62.6 | none | n/a | n/a | Pay per use | API key | Local | ||
|
Inworld's Realtime API runs spoken conversations over WebSocket or WebRTC, chaining Inworld STT, an LLM chosen by the developer and Inworld TTS, with turn detection, interruptions and tool calling. It follows OpenAI's Realtime protocol. Top strength Follows OpenAI's Realtime events, with a migration guide that changes only the URL, key and a few defaults Top weakness No per-minute price. A session's cost depends on STT, LLM tokens and TTS characters, and interrupted segments are still billed |
||||||||||||
| 693 | Rime PrismRime Labs · HTTP API | Voice agents | B | 62.2 | none | n/a | n/a | Freemium | API key | Hosted | ||
|
Prism is Rime's speech-to-speech model for voice agents. It takes caller audio over one WebSocket, decides when the caller has finished, calls the developer's tools and replies in speech with matching text, in English. Top strength Errors page lists 41 codes, each with scope, the field at fault, a correlation ID and what to do Top weakness Rime's status page has no Prism component, so Prism has no incident history of its own |
||||||||||||
| 714 | Ultravox Realtime APIUltravox (Fixie.ai) · HTTP API | Voice agents | C | 61.7 | 3.0 (2) | n/a | n/a | $100 / mo | API key | Hosted | ||
|
Hosted voice agents on Ultravox, an open-weight model that takes speech directly with no speech-to-text step and answers through a TTS voice. Top strength Flat $0.05 a minute including model and built-in voices Top weakness No MCP server |
||||||||||||
| open | MiniCPM-o 4.5OpenBMB · Open-weight model | Open weights | C | 61 | 2.8 (4) | n/a | n/a | Free | None | Local | ||
|
MiniCPM-o 4.5 is OpenBMB's open-weight omni-modal model of 9.4B parameters that takes images, video, audio and text and answers in text and speech, including full-duplex live conversation and voice cloning from a reference clip, under Apache 2.0. Top strength OpenBMB's 4-bit GGUF and Ollama builds fit in about 7 GB Top weakness Speech output and full duplex need trust_remote_code with Transformers 4.51.0 or OpenBMB's llama.cpp-omni |
||||||||||||
| 895 | PlaceCallVOYGR Tech, Inc. · HTTP API | Voice agents | C | 57.2 | 3.5 (4) | n/a | n/a | $0.10 / call | OAuth or key | Hosted | ||
|
PlaceCall, from VOYGR, is an API and remote MCP server for outbound calls to US businesses. An AI voice agent carries out a plain-language task, such as a booking or quote, and returns the outcome and transcript. Top strength Idempotency-Key on POST /calls replays the stored response, with documented 409 answers for a changed or in-flight request Top weakness No status page, incident history, SLA or API changelog found |
||||||||||||
| 928 | Gemini Live APIGoogle · HTTP API | Voice agents | C | 56.6 | none | n/a | n/a | $14 / 1k req | API key | Hosted | ||
|
Google's Live API runs real-time spoken conversations with Gemini audio-to-audio models over a stateful WebSocket. It takes audio, images and text, speaks back, and supports interruptions, function calling and Google Search grounding. Top strength
Top weakness The WebSocket guide authenticates with the API key as a |
||||||||||||
| 934 | Hume EVI (Empathic Voice Interface)Hume AI · HTTP API | Voice agents | C | 56.5 | 2.0 (2) | n/a | n/a | $14 / mo | OAuth or key | Hosted | ||
|
Hume's hosted voice-agent service, accessed through a WebSocket API. Top strength Speech-language model that reads the caller's tone and answers with matching prosody Top weakness Hume ends access to the TTS and EVI APIs on 13 November 2026 and deletes account data after that date |
||||||||||||
| 939 | AssemblyAI Voice Agent APIAssemblyAI · HTTP API | Voice agents | C | 56.4 | 3.5 (4) | n/a | n/a | Pay per use | API key | Hosted | ||
|
AssemblyAI's Voice Agent API runs a spoken conversation over one WebSocket, combining its own speech-to-text, language model and text-to-speech. Agents are stored through a REST API and reached from a browser, a server or an inbound Twilio SIP number. Top strength AsyncAPI 3.0 file for the socket, an OpenAPI file for the token endpoint, and a Markdown copy of every Voice Agent docs page Top weakness No Voice Agent API price on the pricing page or in the billing docs on 2 November 2026, and the free credit no longer names it |
||||||||||||
| 944 | Vocily AIVocily AI Private Limited · HTTP API | Voice agents | C | 56.4 | none | n/a | n/a | $8 / mo | API key | Hosted | ||
|
Vocily AI is a hosted voice-agent platform from Bengaluru for phone calls, WhatsApp and website voice and chat, aimed at Indian languages. Its REST API builds and publishes agents, places outbound calls and reads back transcripts, recordings and costs. Top strength OpenAPI 3.1 file with 54 operations, plus llms.txt and a Markdown copy of every docs page Top weakness No status page, incident history or SLA was found |
||||||||||||
| open | Ming-3.0-audio-tinyinclusionAI (Ant Group) · Open-weight model | STT | D | 52.1 | 2.5 (4) | n/a | n/a | Free | None | Local | ||
|
Ming-3.0-audio-tiny is inclusionAI's open-weight audio-language model for Chinese and English speech recognition, audio understanding and spoken replies, 9.3B parameters with 2.7B active, published on Hugging Face under MIT. Top strength MIT weights in safetensors, downloadable with no account Top weakness No named runtime runs it in a released version, only SGLang-Omni main and inclusionAI's own branches |
||||||||||||
| 1119 | CueLightsshah03 (individual developer, name not stated) · SDK + MCP | Voice agents | D | 51.6 | none | n/a | n/a | Free · OSS | None | Local | ||
|
CueLight is an open-source Python library and set of open-weight models that estimate, every 20 ms of streaming audio, whether a voice-agent user has finished their turn. It runs locally from audio alone, in English. Top strength Apache-2.0 code and weights, with every training dataset and teacher model named in NOTICE Top weakness English only, and the author does not plan other languages soon |
||||||||||||
| 1230 | Synthflow API + MCPSynthflow · HTTP API | Voice agents | D | 47.3 | 2.0 (2) | n/a | n/a | Paid | OAuth or key | Hosted | ||
|
Enterprise voice-agent platform built around a no-code Flow Designer and prompt builder, with a REST Platform API and a hosted MCP server. Top strength Changelog posts on ten days between 9 and 30 October 2026, with API Changes and Deprecation Notices categories and an RSS feed Top weakness Annual enterprise contract only, with no free plan, trial or rate by the minute |
||||||||||||
| open | Qwen3-Omni-30B-A3B-InstructAlibaba (Qwen team) · Open-weight model | Open weights | D | 46.8 | 2.0 (4) | n/a | n/a | Free | None | Local | ||
|
Qwen3-Omni-30B-A3B-Instruct is Alibaba's open-weight omni-modal mixture of experts, a Thinker that takes text, images, audio and video and a Talker that answers in text or speech, about 35B parameters with 3B active, published in BF16 under Apache 2.0. Top strength Takes text, images, audio and video in one prompt and answers in text or speech, in 10 speech output languages Top weakness No quantised build from Alibaba; its only build is 70.5 GB of BF16, about 85 GB to run |
||||||||||||
| 1248 | Bolna API + MCPBolna · HTTP API | Voice agents | D | 46.5 | 2.5 (2) | n/a | n/a | Pay per use | API key | Hosted | ||
|
Voice-agent platform built for Indian languages and phone campaigns. Top strength Transcriber, LLM and voice chosen per agent, or one OpenAI Realtime or Gemini Live model Top weakness Unsigned webhooks and tool calls, IP allowlisting only |
||||||||||||
| 1264 | Vogent APIVogent · HTTP API | Voice agents | E | 45.9 | 2.0 (2) | n/a | n/a | Pay per use | API key | Hosted | ||
|
Phone voice-agent platform with its own conversational LLM, voices and IVR-navigation models alongside GPT. Top strength IVR detection and navigation models for outbound calls Top weakness No public changelog and no release found since January 2026 |
||||||||||||
| 1265 | Vaani Developer PlatformVaani AI Research Private Limited · HTTP API | Voice agents | E | 45.9 | none | n/a | n/a | $10 / mo | API key | Hosted and local | ||
|
Vaani is a hosted voice-agent platform from Bengaluru for outbound and inbound phone calls, browser WebRTC sessions and calling campaigns. Agents reach it through a REST API, a Python SDK and a local MCP server. Top strength Docs as llms.txt and llms-full.txt, with request examples in cURL, Python and JavaScript for each endpoint Top weakness API keys reach the whole workspace, with no scoped, read-only or per-agent keys |
||||||||||||
| open | Ming-3.0-audio-flashinclusionAI (Ant Group) · Open-weight model | STT | E | 44.8 | 2.3 (4) | n/a | n/a | Free | None | Local | ||
|
Ming-3.0-audio-flash is inclusionAI's open-weight audio-language model for Chinese and English. It takes audio with text instructions and returns transcripts, answers about the audio, or spoken replies, 126B parameters with 6.5B active, under MIT. Top strength MIT with no use policy, downloadable with no account or gate Top weakness 259.3 GB of BF16 files and no quantised build, so it needs a multi-GPU server and doesn't fit 128 GB |
||||||||||||
| 1348 | Harmony.aimonday.com Ltd. (Harmony.ai) · HTTP API | Voice agents | E | 41.8 | 2.3 (4) | n/a | n/a | $250 / mo | API key | Hosted | ||
|
Harmony.ai is monday.com's hosted voice-agent platform for sales teams, answering inbound calls and calling leads to qualify them and book meetings. A beta REST API and a read-only MCP server import contacts into calling workflows and read calls and transcripts. Top strength Public OpenAPI 3.0 file, llms.txt and a full Markdown copy of the docs Top weakness The API is in beta, and the docs say operations can change in breaking ways |
||||||||||||
| 1512 | fone.meFone · Platform | Voice agents | F | 16.4 | none | n/a | n/a | $9.99 / mo | None | Hosted | ||
|
fone.me is a hosted AI agent at a personal link that answers calls and messages for a person or small business over web chat, voice, WhatsApp, Telegram and a Premium number. Other agents reach it through keyless MCP and A2A. Top strength Keyless MCP server and per-link A2A endpoints let any agent find, look up and message a person's or business's agent without an account or payment Top weakness No tool arguments, input schemas, annotations, error shapes or rate limits are published for the MCP server or A2A endpoints |
||||||||||||
| 1540 | Fresha AI ConciergeNot for agents Fresha · Platform | Voice agents | F | 10.4 | 1.8 (4) | n/a | n/a | Paid | OAuth or key | Local | ||
|
Fresha AI Concierge answers client calls and messages for salons, spas and clinics. It uses the business account’s services and availability to book, reschedule or cancel appointments and answer questions. Top strength Uses the business's existing Fresha services and calendar availability Top weakness No API, MCP server or other interface for an outside agent |
||||||||||||
Nothing matches these filters. .
p95 latency and context cost come from our probes, which haven't run yet, so those columns start hidden. Grades run from AA to F, and agent-ready means BB or better. Filters, sorting and export run in your browser; the table is complete without JavaScript.
Indexed, not reviewed (1)
Listings sorted into this category from public catalogues (the official MCP registry, APIs.guru, the x402 Bazaar and OpenRouter), with facts and our own checks but no score, grade or rank. How the index works.
| Listing | Kind | What it does | Why it's here |
|---|---|---|---|
| Wixzel Voice wixzel.com | MCP server | Lets Claude, Cursor or any MCP client build voice agents and place real phone calls. | vendor's own |
How we test this category
The same booking and support tasks on every platform. We measure response latency, how interruptions are handled, tool-call success, recovery after a misunderstanding and the total cost per completed call, and record whether the product runs a direct speech-to-speech model or an STT, model and TTS pipeline, since those are different setups.
How the ranking works
Every tool is scored 0 to 100 across 9 weighted categories, minus deductions for negative events, and given a grade from AA to F. Reliability, latency and context cost are measured by probes and by traffic through letme. Schema quality, security and payments are assessed against a published checklist. Task success comes from a fixed task suite. Payments, including x402, are one category of the nine. Read the methodology.
For companies
Do agents find, use and choose your tools?
An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.