Speko Launches Voice-AI Gateway to Unify Fragmented Speech Stacks
The YC S26 startup provides a single API to route STT, LLM, and TTS providers based on real-time performance benchmarks.
Beknazar Abdikamilov, a former Amazon employee and YC S26 founder, has launched Speko to solve the integration friction inherent in building production voice agents. The platform acts as a "Voice-AI Gateway," effectively serving as an OpenRouter for the voice ecosystem.
Speko provides a unified API that encompasses the entire voice stack: Speech-to-Text (STT), Large Language Models (LLMs), and Text-to-Speech (TTS). By consolidating these layers, the platform allows developers to switch between different providers without the need for entirely new integrations. To determine the best routing, Speko benchmarks 61+ speech and language models across 10+ languages, tracking critical performance metrics including Word Error Rate (WER), finalize latency, time to first token, and cost per minute.
The Integration Burden
Building a production-ready voice agent typically requires an ensemble of three distinct models. Historically, developers have selected a specific stack and remained with it due to the high technical friction of re-integrating new vendors. This inertia often persists even as newer, cheaper, or more accurate models enter the market. Abdikamilov founded Speko after spending four years manually benchmarking voice stacks for enterprises across Asia, identifying a systemic need for a decoupled architecture.
Industry Implications
By decoupling the application layer from specific model providers, Speko reduces the ongoing R&D burden of maintaining a voice AI stack. Companies can maintain state-of-the-art performance and cost-efficiency without performing manual migrations every time a new model is released. The impact is most visible for high-volume users; one customer team running thousands of calls per day noted that they can now switch models via a dashboard and have the system handle the transition automatically.
Deployment and Flexibility
To accommodate different security and infrastructure needs, Speko offers an open-source gateway under the MIT license. This is provided as a Go binary that can be deployed as a sidecar, allowing teams to avoid additional network hops or manage their own API keys internally.
What's Next
As the landscape of multimodal models evolves, the industry is watching whether unified gateways can keep pace with native "omni" models that combine speech and text into a single process. For now, Speko focuses on optimizing the existing modular stack through continuous benchmarking and automated routing.