We already have Vapi or Retell working.

Working at what volume? Orchestration is great for prototypes. At production scale you hit latency ceilings, compounding reliability across five vendors, and answer-rate problems from inherited attestation. When you hit that ceiling, the architecture is what changes it.

Can we keep our own models?

Yes. The pipe is the product; the models are your choice. Route to OpenAI, Anthropic, open source or your own fine-tunes. Premium STT and TTS engines route through the same platform.

What does migration actually take?

Numbers port in hours and free from the US and Canada. TwiML runs unchanged on TeXML. Most teams run the carrier swap on a subset of traffic first and expand as it proves out. Typical time to production is weeks, not quarters.

Is HIPAA charged as an extra?

No. HIPAA and a BAA are available on every plan, alongside SOC 2 Type II, PCI, GDPR and ISO 27001, under one DPA.

Contact center

AI contact center infrastructure, owned end to end.

Q: Do you replace our CCaaS?

Not necessarily, and usually not first. Bring us in as the carrier under Talkdesk, UJET or Genesys today and nothing above it changes. Several platforms in this category already run on our network underneath.

AI handles intake, humans handle escalation, and the caller never hears the seam. Voice AI agents, speech and telephony on one private network, under one contract.

Book a demo Why calls get answered

Countries, licensed carrier: 45+
Attestation level, signed at origin: A
Continents with GPUs: 4
Bill: 1

14,000+ industry-leading companies choose Telnyx

01 — Ownership

Answered, not spam filtered.

On a multi-vendor stack, 30 to 40% of outbound calls never ring: attestation is inherited from whoever owns the trunk, and carriers filter what they cannot verify. We sign calls as the originating carrier, the highest trust level the network supports.

Telnyx, the originating carrier. Attestation A

Your contact centerTelnyx, the carrierAttestation A, signed at origin

20 to 30% higher answer rates against a multi-vendor baseline.

Stitched stack: voice AI platform + CPaaS + carrier. Attestation B or C

Your contact centerVoice AI platformCPaaSUnderlying carrierAttestation B or C, inherited

Each connector on the stitched path is a vendor boundary; trust can only be passed on, never raised. For example Vapi or Retell on Twilio.

Only the originating carrier can sign A-level. We are the originating carrier.

02 — Physics

500 ms per turn. Spend it on the answer.

STT, LLM and TTS run on GPUs we own, in the same racks as the media plane, with the agent on Edge Compute beside them. Network round trip stays under 200 ms, so nothing crosses a vendor between hearing and speaking.

Telnyx architecture, one turn

0 vendor boundaries0 internet transitsattestation A

Caller speaksPSTN

Stateful actorAgentSDK

STTInferenceTTS

Business logic

PSTNCaller hears

<500 ms latency

Overall architecture. Call audio runs PSTN, STT, Inference, TTS, PSTN on one line; AgentSDK orchestrates it from the Stateful actor runtime, with business logic beside it.

03 — Composition

Pick a use case and watch the blocks light up.

The same primitives, composed per job, on one control plane. Five vendors taking margin becomes one that owns the stack; typical savings run 40 to 60% at scale, a consequence of ownership rather than a promotion.

CHOOSE BY USE CASE OR PRODUCT

CommsVoice API
CommsSIP Trunking
CommsTeXML
CommsGlobal numbers
AIVoice AI Agents
AIText-to-Speech
AISpeech-to-Text
EdgeStatefulActor
EdgeInference
EdgeAgentSDK
EdgeObject Storage
IntegrationZendesk

RESULTS

cURL

# intake and triage — create the agent# same endpoint for every workloadcurl -X POST https://api.telnyx.com/v2/ai/assistants \  -H "Authorization: Bearer $TELNYX_API_KEY" \  -H "Content-Type: application/json" \  -d '{    "name": "intake",    "model": "meta-llama/Meta-Llama-3.1-8B-Instruct",    "instructions": "Answer, triage, summarize. Escalate to a person with the transcript attached."  }'# 11 primitives · 1 vendor · 1 bill

Intake and triage

Primitives composed: 11 of 50
Est. latency budget: ~460 ms
Est. unit cost (all-in): $0.05/min
Vendors required: 1

SELECT USE CASE

OR SELECT PRODUCT

IntegrationCCaaS / CRM

RESULTS

cURL

# intake and triage — create the agent# same endpoint for every workloadcurl -X POST https://api.telnyx.com/v2/ai/assistants \  -H "Authorization: Bearer $TELNYX_API_KEY" \  -H "Content-Type: application/json" \  -d '{    "name": "intake",    "model": "meta-llama/Meta-Llama-3.1-8B-Instruct",    "instructions": "Answer, triage, summarize. Escalate to a person with the transcript attached."  }'# 11 primitives · 1 vendor · 1 bill

Intake and triage

Primitives composed: 11 of 50
Est. latency budget: ~460 ms
Est. unit cost (all-in): $0.05/min
Vendors required: 1

RESULTS

cURL

# intake and triage — create the agent# same endpoint for every workloadcurl -X POST https://api.telnyx.com/v2/ai/assistants \  -H "Authorization: Bearer $TELNYX_API_KEY" \  -H "Content-Type: application/json" \  -d '{    "name": "intake",    "model": "meta-llama/Meta-Llama-3.1-8B-Instruct",    "instructions": "Answer, triage, summarize. Escalate to a person with the transcript attached."  }'# 11 primitives · 1 vendor · 1 bill

Intake and triage

Primitives composed: 11 of 50
Est. latency budget: ~460 ms
Est. unit cost (all-in): $0.05/min
Vendors required: 1

One platform. One bill. No fingerpointing.

04 — Proof

Running in production, at scale.

Replicant
3 s to under 1 s
Talk time latency, with voice cost down 86%. Hundreds to hundreds of thousands of calls a day on the same media path.
Read the story
2 weeks to integrate
Down from two to three months. Voice API for voice biometrics and SIP for the agent copilot.
Read the story
Ensurity Group
20M calls a year
Across 14 companies, up from 15 to 18 million, on SIP trunking, Number Lookup and Voice AI Agents.
Read the story

Replicant
3 s to under 1 s
Talk time latency, with voice cost down 86%. Hundreds to hundreds of thousands of calls a day on the same media path.
Read the story
2 weeks to integrate
Down from two to three months. Voice API for voice biometrics and SIP for the agent copilot.
Read the story
Ensurity Group
20M calls a year
Across 14 companies, up from 15 to 18 million, on SIP trunking, Number Lookup and Voice AI Agents.
Read the story

05 — Migration

You don't have to replace your CCaaS to start.

The common path is not a rip and replace. We land underneath what you already run, prove quality and cost on live traffic, and the AI layer moves when there is a reason to.

01Carrier first
We sit under Talkdesk, Genesys, UJET or your own stack as the SIP carrier. Nothing above changes; the rate and attestation do.
02Port the numbers
Hours, not weeks, free from the US and Canada, in programmatic batches.
03TwiML runs unchanged
Point existing applications at TeXML. This is the step that removes the rewrite from the business case.
04AI intake last
Move intake, transcription and summarization onto our GPUs when the latency argument has been proved on your own traffic.

“Telnyx has structured their API in a way that makes it very easy to move from other similar providers in the space.”

Dave Bullock, Chief Technology Officer, UJET

Read the story

06 — Compliance

One DPA. One audit. Not five.

We are a licensed carrier in 45+ countries, regulated as communications infrastructure rather than a SaaS vendor. Voice data, including PII and PHI, stays inside one network boundary. HIPAA and a BAA on every plan, and in-region processing by architecture rather than configuration.

SOC 2 TYPE II
HIPAA + BAA
PCI
GDPR
ISO 27001

What a security review asks

Who signs the call, and at what attestation level
Where do recordings and transcripts physically sit
Which model vendors see customer audio
One vendor to assess, or five

Hear the difference

Book a demo on your own traffic, or start on the API today. Every rate is published.

BOOK A DEMO SIGN UP

FAQ

Not necessarily, and usually not first. Bring us in as the carrier under Talkdesk, UJET or Genesys today and nothing above it changes. Several platforms in this category already run on our network underneath.

AI contact center infrastructure, owned end to end.

Running in production, at scale.

3 s to under 1 s

2 weeks to integrate

20M calls a year

3 s to under 1 s

2 weeks to integrate

20M calls a year

You don't have to replace your CCaaS to start.

One DPA. One audit. Not five.

What a security review asks

Hear the difference

FAQ

Do you replace our CCaaS?

We already have Vapi or Retell working.

Can we keep our own models?

What does migration actually take?

Is HIPAA charged as an extra?

Ask AI