Blog

jambonz now supports OpenAI's GPT Live, a full-duplex speech-to-speech model. Connect a phone number and start from working code today.

xAI's Grok Voice Think Fast 2.0 now works with jambonz as a native speech-to-speech vendor. Connect phone calls with one verb and a reasoning-effort dial.

Add live call monitoring, agent coaching, and barge-in to any conference with jambonz v11's conference listen fork. Open-source reference app included.

Connect phone calls to Qwen-Audio-3.0-Realtime with a single verb: full-duplex voice AI with semantic turn detection, streaming transcripts, and sub-second response latency.

jambonz v11 adds built-in blind, warm, and conferenced handoff for voice agents: declarative, with no custom transfer workflow to build.

jambonz v11 replaces FreeSWITCH with a custom media server, more than doubling capacity and improving voice AI platform performance.

The AI dev community runs on Python — so we built a Python SDK for jambonz. A tour of the verbs, plus a webhook IVR and a WebSocket voice AI agent.

jambonz introduces a commercial license option to fund faster innovation — and, in return, to stop making self-hosting deliberately hard.

A deep dive on jambonz LLM and TTS streaming for low-latency Voice AI — the websocket protocol, the say verb, the Node.js SDK, and a full Anthropic example app.

Step-by-step: deploy jambonz as the custom telephony layer for Retell AI, covering both elastic SIP trunking and dial-to-SIP-endpoint, plus outbound calls.

I added OpenAI's Realtime API to jambonz, then ran the caller's audio through Deepgram in parallel to compare accuracy and latency — and what that reveals about speech-to-speech vs. traditional ASR pipelines.

How SoundOfData and Callable.io set up jambonz as a Direct Routing SBC for Microsoft Teams — drachtio TLS, jambonz settings, and Microsoft 365 carrier and customer tenant configuration.

Bring your own carrier to Retell AI with jambonz and drop the Twilio bundle — carrier setup, phone numbers, the streaming shim app, and call transfer.

We benchmarked time-to-first-byte across seven TTS vendors — Deepgram, ElevenLabs, Google, Microsoft, PlayHT, Rime Labs, and Whisper — for conversational AI. Here's the jambonz TTS latency leaderboard.

Using an LLM to predict what kind of answer a caller will give, then tuning the speech recognizer's endpointing for that turn — an experiment in better turn detection.

jambonz 0.8.4 adds a priority attribute to queued calls, so urgent callers can jump the queue — with a Node-RED example flow.

Mirror SIP and RTP from your jambonz SBCs to a Voipmonitor server with AWS Traffic Mirroring — plus a bonus: rolling pcaps uploaded to S3.

Conversational AI has unique needs that long-form transcription doesn't — endpointing control, prompt-aware APIs, and pause/resume billing. Three features speech vendors still aren't shipping.

Build a jambonz image on GCP with packer, then stand up a VM with terraform — service account, image build, terraform variables, DNS, and first login.

The jambonz custom speech API lets you plug in any STT engine via a websocket server. Here we wire up Vosk, an open source recognizer you can run yourself.

A walkthrough of deploying jambonz from the AWS Marketplace CloudFormation offering, plus the post-install steps to move the portals from http to https.

Add SIP over secure websockets to a jambonz install: a Let's Encrypt wildcard cert, a drachtio wss contact on 8443, and the SIP realm and registration webhook in the portal.

Scaffold a jambonz webhook application with create-jambonz-app, walk through the generated Express code, and wire it up to a phone number.