We’re partnering with Gradium to join realtime voice AI and web search. Gradbot, Gradium’s open source agent framework for fast, natural conversations is adding support for Keenable search. Ask any voice agent something it has to look up, and you can hear the moment it happens: silence, while it searches. With Keenable under the hood, a Gradbot agent pulls answers from the live web as naturally as it speaks from memory. With Gradium’s voices, Keenable search experience now benefits from natural conversation and great voices. Hear it in the hosted demo: gradium.keenable.ai.
The latency problem with open web
Voice agents are good at things that live in a model’s weights. Some facts change too fast to ever be in the weights: this morning’s headline, today’s price, whether the flight is delayed. Others were never there at all: the API quirk of a niche library, the opening hours of a restaurant with twelve reviews, the name of the second-largest employer in a small town. Fresh or rare, the failure is the same, the model either guesses or goes looking. Most search APIs were built for batch use, not for a conversation waiting on the answer, so a single lookup runs into seconds and blows the latency budget.
That leaves a bad choice: keep the voice instant and let the answer be wrong, or fetch the real answer and let the pause break the conversation. Either way it stops being a conversation and becomes a query box with a voice on top.
The stack
Gradium builds low-latency streaming STT and TTS with prosody that holds up in live, back-and-forth speech; time to first audio stays inside the conversational budget. Keenable returns web retrieval in under 200ms and is fast enough to sit inside the agent’s reasoning loop instead of interrupting it. Ask the agent whether your train is on time, and it checks before it finishes drawing breath.
Fast voice, realtime retrieval, on the same budget.
Hear it for yourself
For this partnership we wired Keenable search into a Gradbot live demo, so the agent pulls from the open web mid-conversation and speaks it back. Ask it something it can’t possibly know (yesterday’s football score, the weather where your mother lives) and listen for the pause that isn’t there.
Hosted demo: gradium.keenable.ai. To add Keenable and Gradium to your own stack, see the integration docs at keenable.ai/integrations and Gradium’s documentation.