{
  "data": null,
  "kind": "anchor.page",
  "links": {
    "api": "https://www.anchorterminal.com/api/v1/index.json",
    "html": "https://www.anchorterminal.com/niches/voice-agents.long-call-session/",
    "json": "https://www.anchorterminal.com/niches/voice-agents.long-call-session/index.json",
    "llms": "https://www.anchorterminal.com/llms.txt",
    "markdown": "https://www.anchorterminal.com/niches/voice-agents.long-call-session/index.md",
    "mcp": "https://www.anchorterminal.com/mcp",
    "slim": "https://www.anchorterminal.com/niches/voice-agents.long-call-session/index.min.md"
  },
  "markdown": "Four or more companies' docs show a full fit.\n\n| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |\n| --- | --- | --- | --- | --- |\n| 4 | 13 | 29 | 7 | 4 |\n\n## What counts as done\n\nThe call continues past 30 minutes with context carried over, and no tool result replayed or re-executed after the reconnect.\n\nMeasured, for our own tests: no audible gap over 1 second.\n\nHow agents ask for it: calls drop at 10 minutes; the session ends after 8 minutes; resume the conversation without the caller noticing\n\n## Does the job\n\nThe docs show it meets the test above, and nothing we read says otherwise.\n\n- **Deepgram Voice Agent API (Deepgram, Inc.)**: Sessions last 2 hours with a MAXIMUM_SESSION_LENGTH_APPROACHING warning 5 minutes ahead; a new session can be seeded with agent.context messages including function calls. KeepAlive every 8 seconds holds idle sockets.\n- **Retell AI API + MCP (Retell AI Inc.)**: max_call_duration_ms defaults to 1 hour and goes to 2 hours, end_call_after_silence_ms defaults to 10 minutes (minimum 10 s); the platform holds the call, so no client reconnect is needed.\n- **Rime Prism (Rime Labs, Inc.)**: The limits table states no conversation time limit; only 5 minutes idle closes it, and context beyond 65,536 tokens drops the oldest turns.\n- **Vapi API + MCP (Vapi Inc.)**: A cascaded pipeline with no provider socket limit exposed; maxDurationSeconds runs 10 to 43,200 s (default 600) and silenceTimeoutSeconds 5 to 3,600 s, so a call can be configured past 30 minutes.\n\n## Does the job, with a limit\n\nIt 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.\n\n- **Amazon Nova 2 Sonic (Amazon Web Services, Inc.)**: Connections end at 8 minutes; continuation means a new session with chat history sent once, implemented by the client from samples. Reconnects can replay tool calls.\n- **AssemblyAI Voice Agent API (AssemblyAI, Inc.)**: session.resume keeps context for 30 seconds after a drop, but the session ends at max_session_duration_seconds with close 1008 and no warning event; behaviour of pending tools across a resume is not documented.\n- **Bland AI API + MCP (Intelliga Corporation)**: A phone call has no socket lifetime to manage, but max_duration defaults to 30 minutes and ends the call; it must be raised per call or agent.\n- **Bolna API + MCP (Voxlabs Private Limited)**: Phone calls have a configurable total call timeout rather than a socket lifetime, but hang-up on user silence (6 to 10 seconds suggested) ended calls at about 12 seconds in a reproduced account.\n- **ElevenLabs Agents API + MCP (Eleven Labs Inc.)**: max_duration_seconds defaults to 600 (10 minutes) and can be set from 60 to 7,200, so a long call needs the setting raised; no reconnect is needed within that limit.\n- **Gemini Live API (Google LLC)**: Connections last about 10 minutes with a GoAway warning; resumption handles valid 2 hours and context compression are documented but off by default. Calls dropped where the client only logged GoAway (acc-15), and Pipecat replayed settled tool results on resume (acc-24).\n- **GPT-Live (OpenAI OpCo, LLC)**: Connected SIP calls last up to 2 hours and context over 90 per cent is summarised into a replacement engine inside the same session. After a drop, continuation is only by forking a finalised stored recording, unavailable under ZDR.\n- **LiveKit Agents (LiveKit Incorporated)**: A pipeline session has no socket lifetime and WebRTC reconnects automatically, and user_away_timeout only marks the user away; with a realtime model plugin, the provider's session limit and rotation depend on the plugin (one documents a 60-minute close with no rotation).\n- **Pipecat (Daily)**: A pipeline session has no socket lifetime, but the Gemini Live service resends every settled tool result to the new session after a resumed reconnect (open issue), and the worker's idle timeout defaults to 300 seconds.\n- **Synthflow API + MCP (AgentFlow AI GmbH)**: With Limit Call Duration off no fixed cap applies (the slider tops out at 30 minutes), and Max Idle Duration is configurable, but its default is not stated and no 30-minute account exists.\n- **Ultravox Realtime API (Theai, Inc. dba Inworld AI)**: maxDuration defaults to 3600 s on one call, which covers 30 minutes; beyond that, priorCallId starts a new call with the old history, which is a new call rather than a smooth reconnect.\n- **Vaani Developer Platform (Vaani AI Research Private Limited (Bengaluru, India, CIN U62099KA2025PTC201212))**: A cascaded pipeline with max_call_duration configurable from 1 to 60 minutes and idle hangup timeouts configurable, so 30 minutes is reachable, but 60 minutes is a hard ceiling and WebRTC sessions default to a 30-minute TTL.\n- **Vogent API (Monoid, Inc.)**: maxDurationSeconds is set per agent with HANGUP or TRANSFER at timeout, and silence hangup can be DISABLED, but no allowed range or default is stated.\n\n## Couldn't check\n\nThe docs were closed to us, or silent. These carry no mark either way.\n\n- **fone.me (Fone)**: No call length limit is stated in any page that loaded.\n- **Fresha AI Concierge (Fresha.com SV Ltd)**: No call length limit is published.\n- **Harmony.ai (monday.com Ltd.)**: No call length limit or reconnect behaviour is documented.\n- **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.\n- **Inworld Realtime API (Theai, Inc. dba Inworld AI)**: The session object has a read-only expires_at but no page read gives a maximum session length or a reconnect and resume procedure.\n- **OpenAI Realtime API (OpenAI OpCo, LLC)**: The guides read state no Realtime session lifetime and document no resumption or reconnect; whether a call passes 30 minutes was not established.\n- **PlaceCall (VOYGR Tech, Inc.)**: No maximum call length or hold time limit is stated in the repository docs, and the API docs are closed to our crawler by robots.txt.\n\n- 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.\n- not this job: Pipecat PhoneLLM (Daily): No session or connection of its own; a text model only.\n- not this job: Vocily AI (VOCILY AI PRIVATE LIMITED): max_duration_minutes has a maximum of 30, so a call cannot continue past 30 minutes.\n\n## What stops agents finishing\n\n- GoAway or 8-minute close with no client reconnect logic\n- resume refused with 1011 after a silent connection loss\n- settled tool results re-sent to the new session\n- inactivity timeout ends the call at 10 seconds of silence\n\n## Around it\n\nBefore: voice-agents.inbound-receptionist. \n\n## Where this comes from\n\nWeight 7 of 100 in Conversational voice agents: 4 of 31 accounts (acc-13, acc-15, acc-23, acc-24)\n\nAs data: https://www.anchorterminal.com/api/v1/use-cases/voice-agents.long-call-session.json, or the MCP tool get_use_case with id voice-agents.long-call-session. Draft, read 2026-11-12 from public docs and first-hand accounts, not yet reviewed by a person.\n",
  "meta": {
    "attribution": "Anchor Terminal (https://www.anchorterminal.com)",
    "attributionRequired": true,
    "credit": "CC BY 4.0, credit Anchor Terminal and link the page or API address you took it from (https://www.anchorterminal.com/terms/#data)",
    "docs": "https://www.anchorterminal.com/docs/",
    "generatedAt": "2026-11-16",
    "license": "CC-BY-4.0",
    "method": "https://www.anchorterminal.com/benchmark/",
    "methodology": "0.8",
    "openapi": "https://www.anchorterminal.com/openapi.json",
    "preview": false,
    "run": "2026-10-01",
    "runLabel": "October 2026 research run",
    "terms": "https://www.anchorterminal.com/terms/#data"
  },
  "page": {
    "breadcrumbs": [
      {
        "name": "Home",
        "url": "https://www.anchorterminal.com/"
      },
      {
        "name": "For agents",
        "url": "https://www.anchorterminal.com/agents/"
      },
      {
        "name": "Niches",
        "url": "https://www.anchorterminal.com/niches/"
      },
      {
        "name": "Keep a call alive past the model's connection limit",
        "url": ""
      }
    ],
    "description": "Keep a call alive past the model's connection limit: 4 companies fully do it, 13 with a limit, 29 carry the tag. Every listing in Conversational voice agents with the reason and a source.",
    "facts": null,
    "h1": "Keep a call alive past the model's connection limit",
    "image": "https://www.anchorterminal.com/assets/og/default.png",
    "path": "/niches/voice-agents.long-call-session/",
    "published": "",
    "section": "agents",
    "title": "Keep a call alive past the model's connection limit | Anchor Terminal",
    "toc": null,
    "updated": "2026-11-15",
    "url": "https://www.anchorterminal.com/niches/voice-agents.long-call-session/"
  },
  "tokens": {
    "markdown": 2000,
    "slim": 430
  },
  "version": 1
}
