Rebar Wiki

elevenlabs-agents

5 min read

ElevenLabs Agents

platform voice-ai telephony hubspot

ElevenLabs’ conversational-agent platform — the natural successor to GoodCall / similar reverse-engineered receptionist services. Real WebSocket + SDK + webhook surface, BYO telephony via Twilio or SIP, native HubSpot integration. Where the ai-receptionist-to-hubspot-bridge pattern can be retired.

Two products, one stack

ProductAudienceUse it for
ElevenAgents (developer platform)Engineers building voice appsCustom agent flows, tool calls, RAG over your docs, programmable voice + chat
Reception by ElevenAgentsSMB ownersTurnkey AI receptionist with dedicated US number. Currently free Alpha, paid tier hinted ~$22/mo

Both are built on the same underlying agent engine; “Reception” is just a packaged config for inbound-call answering with scheduling, FAQ handling, and basic CRM hooks pre-wired.

Why it kills the polling pattern

Compared to ai-receptionist-to-hubspot-bridge’s polling-based GoodCall flow:

Polling-based bridgeElevenAgents
Reverse-engineered cookie/JWT, ~60d rotationStandard API keys
No webhooks → 5-minute cron + lookback overlapPost-call webhooks → push, not pull
Custom Bun + Hono service on RailwayJust a webhook receiver (~50 LOC) or no service at all if using native HubSpot integration
Phone format dedup + 10-variant searchTool calls fire your endpoints with structured data inline
Pagination + 100 calls/run capReal-time, no polling cap
Tier-restricted scope workaround required (legacy v1 engagements API)HubSpot integration is native; field mapping is config

If you’re authoring a new integration and have a choice, start here. The reverse-engineered pattern is for when you’re stuck with a platform that won’t expose webhooks.

Pricing

Per-minute pricing on every paid tier; included minutes scale with the plan:

Plan$/moMinutes includedConcurrent calls
Free$0154
Starter$6756
Creator$2227510
Pro$991,23820
Scale$2993,73830
Business$99012,37540

Overage = 0.16/min. LLM tokens billed separately — bring your own OpenAI/Anthropic key, or use ElevenLabs-managed.

For SMB volume — say a master electrician fielding 200 calls/month at ~3 min each = 600 min — that’s 54/mo** plus LLM. Well under the Pro tier; Starter or Creator + small overage is the sweet spot.

Telephony — three paths to a phone number

1. Twilio native (simplest)

Buy a number in Twilio (~2 to validate end-to-end.

2. SIP trunk (port your existing number)

Compatible with Twilio Elastic SIP, Telnyx, Vonage, RingCentral, Plivo, Bandwidth, Sinch, Infobip, Exotel — and most standard SIP providers. Use this when you want to keep an existing business line without porting it to Twilio. Audio is G.711 (8kHz) or G.722 (16kHz). TLS transport + SRTP supported and recommended for prod.

3. Reception by ElevenAgents (turnkey)

Skip Twilio entirely. ElevenLabs assigns a dedicated US number on signup. Less control (no custom telephony events, no SIP), but zero infra. Currently US-only.

Effective per-call cost

Stacked all-in for a typical inbound call:

LayerCost
ElevenLabs agent$0.08/min
Twilio number rental$1.15/mo flat
Twilio inbound voice~$0.0085/min (US local)
LLM tokens~$0.001–0.005/min depending on model

**Effective: ~1.15/mo per number.

Integration surface

Documented

  • Documented platform with developer SDKs (JavaScript, React, React Native, Python, iOS/Swift, Kotlin, Flutter)
  • WebSocket API + WebRTC connection types (WebRTC = lower latency, recommended for prod)
  • Embeddable widget (<elevenlabs-convai>) for in-page web/voice
  • Mid-conversation tool calls (agent fires your endpoints with structured args; result returns to the conversation)
  • Post-call webhooks (push the conversation_id, transcript, recording URL, collected variables when a call ends)
  • Conversation list/export API: GET /v1/convai/conversations, GET /v1/convai/conversations/{id}
  • Server-side conversation tokens for low-latency WebRTC (no client-side API key exposure)

Native CRM/automation integrations

HubSpot, Salesforce, Zendesk, Stripe, Cal.com, n8n. For a HubSpot-centric stack, the agent can be configured to upsert contact + log Call activity natively — no custom service needed for the standard flow.

Compliance

ISO 27001 + ISO 42001 certified. No HIPAA support yet — flag if any client work is medical-adjacent.

Migration path from a polling-based bridge

If you currently run ai-receptionist-to-hubspot-bridge against an upstream like GoodCall, the migration is:

  1. Spin up an ElevenAgents account and configure the agent with your existing FAQ + form questions as data-collection variables.
  2. Wire telephony. Either point a new Twilio number at the agent (cleanest) or repoint an existing SIP trunk.
  3. Configure the post-call webhook to fire at your endpoint. Payload includes conversation_id, transcript, collected_variables, recording_url.
  4. Replace the polling Bun service with a thin webhook receiver (Hono on Railway, ~50 LOC) that maps the structured payload to your existing HubSpot custom properties (preserves chronological <platform>_call_log history). OR drop the receiver entirely and use the native HubSpot integration if its field mapping is sufficient.
  5. Cut over by forwarding the original business line to the new ElevenLabs number.

End state: no JWT rotation, no Cloudflare 429s, no 60-day token rotations, no scope workarounds, fewer moving parts.

When ElevenAgents is the wrong choice

  • HIPAA-required workloads — not yet certified.
  • Non-US numbers, day one — Reception is US-only at present, though the dev platform supports BYO international SIP. Verify country coverage before quoting.
  • Hyper-domain-specific voice quality requirements — some industries (medical dictation, legal stenography) need specialty providers; ElevenLabs is general-purpose conversational.

Graph View

Backlinks