MEDIA STREAMING

Bring your model. We'll stream the call straight to it.

Teler opens a WebSocket to your server and streams live call audio both ways - plug in any AI model; no telephony code required.

Right Column (6)

How it works

One WebSocket, both ways. One flow endpoint to tell Teler what to do. Nothing else to build.

The call connects

Someone calls your Teler number, or your code dials out. Teler hits your flow_url, and your server tells it to open a WebSocket for live audio.

The WebSocket opens

Your server holds a full-duplex WebSocket. The caller’s audio arrives as PCM L16 frames; your agent’s audio goes back down the same socket.

Your voice pipeline handles the intelligence

Speech-to-text, your model, text-to-speech- any provider, or your own. This part is yours. We stay out of it.

Your agent replies

Teler plays the audio into the live call, handling jitter and sync. Your code can interrupt the agent, hand off to a person, or hang up mid-call.

Built for AI voice agents

Interrupt your agent, instantly
Asset 14 (4)

Interrupt your agent, instantly

When your system detects the caller talking, tell Teler to clear the queued audio - your agent goes quiet right away instead of talking over them.

Asset 14 (5)

Hand off to a human

When the caller asks for a person, tell Teler to transfer the call to a colleague. Set what they hear, like a hold tone, while it connects.

Asset 14 (6)

Under 100ms, each way

Audio reaches your server and comes back in under 100ms. Fast enough that the caller never notices a gap.

Asset 14 (7)

Every event replayable

Every event and recording is signed and securely stored. If your endpoint misses a webhook, you can request it later through the API.

What you can build

Route calls without a menu

Your agent asks what the caller needs and transfers them to the right team. No more “press 1 for billing.”

Catch an upset caller early

Check sentiment on the live audio and hand the call to a person the moment the caller sounds upset.

Remember every caller

Teler sends the caller’s number when the call starts, so your agent can pick up from the last call.

Stay compliant on every call

Check what both sides say in real time. Play a required disclosure, or end the call, the moment a rule is hit.

Verify callers before you help

Run your own voice-ID or fraud check on the live audio, then transfer or end the call if something looks off.

Speak the caller’s language

Plug in Sarvam or Deepgram to detect the caller’s language, even when they mix two, and reply in it.

Works with your favorite tools

Teler connects natively with OpenAI, ElevenLabs, Gemini, Vapi, Deepgram, Ultravox, and any custom model - no lock-in, and you pay your provider directly.

Frame 2147257698 (1)

What our customers say about us

Frequently Asked Questions

Have more questions?
Reach out to us at sales@frejun.com.

What latency can I expect?

Audio reaches your server and comes back in under 100ms each way - fast enough that callers don't notice a gap.

Any of them. Teler works with OpenAI Realtime, Gemini Live, ElevenLabs, Ultravox, Vapi, and Deepgram - or bring a fully custom pipeline.

Caller audio reaches your server as PCM L16, at 8kHz or 16kHz. Audio you send back to the call needs to be 8kHz. The reference bridges on GitHub handle the resampling for you.

No. Teler carries the audio and the call controls. The listening and deciding happen in your own software, with whatever models you choose.

No. Once a call is transferred, the AI agent steps out of the conversation. Teler continues to manage the call, but the AI agent’s monitoring and analysis cover only the portion of the conversation it handles.

Every event is signed and securely stored, so you can retrieve anything your server missed through the API.

Both. Use the SDK to move fast, or build directly against the WebSocket protocol for full control.

Connect your model to its
first call today

Get a number, connect your model, and take your first call in minutes.