AI VOICE AGENTS
Put your AI voice agent on a phone number
Teler gives your agent a phone number and streams the call audio to your server, both ways, live. Your model, your prompt, your voice.
We take the telephony.
You keep the agent.
We Handle
- Carriers, numbers and routing
- Live audio streaming, both directions
- Clearing queued audio on request
- Signed events and call recordings
You own
- Your model and your prompt
- Speech-to-text and text-to-speech
- When to interrupt
- What to store, and for how long
How it works
The whole loop, from the call connecting to your agent’s reply.
The call connects
Your code dials out, or someone rings your Teler number. We request your flow_url with the session details, and your response tells us what to do with the call.
The audio stream opens
Your server holds a full-duplex WebSocket. Raw caller audio arrives as L16/8000Hz frames; your agent’s audio goes back down the same socket. Both directions, whole call.
Your agent works out what to say
Speech to text, your model, text to speech — any vendor, or your own. This part is yours and we stay out of it, which also means the thinking time is yours to optimise.
Your agent replies
Teler injects the audio into the live call and handles jitter buffering and sync. Interrupt it, hand the call to a person, or hang up — any time, mid-call. Every event is logged and the recording is stored.
Built for AI voice agents
Stop your agent mid-sentence
Hand off to a human
Under 100ms, each way
Every call replayable
Stop your agent mid-sentence
When the caller cuts in, tell Teler to stop. Your agent goes quiet instantly instead of talking over them.
Hand off to a human
When the caller asks for a person, pass the live call to a colleague. You set what each hears.
Under 100ms, each way
Audio reaches your server and comes back in under 100ms. Fast enough that the caller never hears a gap.
Every call replayable
Every event and recording is signed and stored; you can request any that your server missed.
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.
Agents you can build
Inbound support
Answer every call instantly, resolve what your agent can, and hand the rest to a person mid-conversation.
Verification and screening
Confirm who you're talking to, screen an applicant, or reach a driver — with the call retrievable months later.
Appointment reminders
Thousands of short outbound calls triggered by your systems, with confirmations written straight back.
Voice inside your product
Give your own customers a voice agent without becoming a phone company. Numbers by API, logic in your repo.
Orders and delivery updates
Confirm an order, give a delivery window, handle a return. Change the script when the campaign changes.
Lead qualification
Call new leads within seconds of a form submission and route the ones worth a human's time.
What latency can I expect?
Teler streams call audio over a persistent WebSocket connection with configurable chunk sizes as small as 20 milliseconds, so you control the trade-off between latency and throughput. In a typical conversational AI setup, transport latency is a small fraction of the total round trip - end-to-end responsiveness is dominated by your AI provider's processing time, not by Teler.
Which AI models can I use?
Teler is model-agnostic. Connect OpenAI, Anthropic, ElevenLabs, Gemini, Vapi, Deepgram, Ultravox, or any custom model - there's no lock-in, and you pay your AI provider directly.
Can I test FreJun Teler without a credit card?
Yes - sign up and get 25,000 free credits to build and test before you ever enter payment details.
How do I get a phone number for my agent?
Provision one instantly via API or dashboard, in 50+ countries - local, toll-free, or mobile numbers.
What happens if my server misses an event?
No event is silently dropped. If your endpoint is unavailable, Teler automatically retries delivery with progressive backoff. Every event is also signed and stored with its complete delivery history, so once your service recovers you can list any missed events through the Events API and redeliver them to your endpoint. Events remain available for retrieval and redelivery for 30 days.
Do you provide SDKs, or do I build directly against the WebSocket?
Both. Official Python and Node.js SDKs handle the REST API and real-time media streaming for you, so you can integrate voice AI in minutes. If you prefer complete control, the REST API and WebSocket media protocol are fully documented for direct integration.
Put your AI voice agent
on a phone number today
Get 25,000 free calling credits on your first upgrade with Teler. Connect your model, add a number and take your first call.