Launch HN: Speko (YC S26) – OpenRouter for Voice AI
Speko, a YC S26 launch, introduces an 'OpenRouter for Voice AI' to address the complexity of selecting optimal speech-to-text, LLM, and text-to-speech models for voice agents. It automates model benchmarking and routing based on custom constraints, aiming to keep voice applications using the best and most cost-effective AI. This appeals to HN's developer-centric audience by tackling a common integration headache with a transparent, API-driven solution and an open-source component.
The Lowdown
Speko (YC S26) has launched on Hacker News, presenting a platform designed to simplify and optimize the selection of AI models for voice agents, branding itself as an 'OpenRouter for Voice AI.' The core problem it tackles is the rapid evolution and multitude of speech-to-text (STT), large language model (LLM), and text-to-speech (TTS) options, which makes continuous evaluation and integration a significant challenge for developers.
- Dynamic Model Selection: Speko provides an API that intelligently routes requests to the optimal combination of STT, LLM, and TTS models based on user-defined criteria such as accuracy, latency, cost, language, and region.
- Continuous Benchmarking: The platform maintains public, impartial benchmarks by consistently testing models with identical inputs. These benchmarks are transparent, even showcasing instances where Speko's selections underperform, and include sophisticated scoring for TTS naturalness.
- Open-Source Gateway: Speko offers an MIT-licensed, open-source gateway that can run as a sidecar, allowing users to avoid extra network hops and manage their own keys (BYOK mode) without sharing them with Speko's cloud. This gateway operates independently after initial setup.
- Business Model: While the open-source gateway and BYOK setup are free, Speko charges for its hosted router service and managed keys, offering consolidated billing.
- Early Traction: The company reports impressive early growth, with external usage increasing by approximately 25 percent weekly since the batch started.
Speko positions itself as an essential tool for companies building voice agents, promising to keep their applications competitive and efficient by ensuring they always leverage the best available voice AI models without constant manual re-evaluation.
The Gossip
Competitive Comparisons & Core Competencies
The discussion immediately turned to how Speko differentiates itself from existing solutions. Users inquired about its distinction from Livekit Gateway and managed platforms like Vapi. The founder clarified that Speko's primary value lies in its continuous, public benchmarking and intelligent routing of voice models based on specific constraints. Additionally, its open-source, self-hostable gateway offers a crucial differentiator by providing greater control and preventing vendor lock-in, contrasting with the fully managed infrastructure of competitors.
Architectural Assumptions & Shifting Paradigms
A prominent critique challenged Speko's fundamental premise that production voice agents are typically an 'ensemble of three models.' One commenter argued that the industry is rapidly moving towards single, end-to-end models for voice AI, driven by better performance and lower latency. They suggested Speko's true value might be in 'automatic evaluations' (akin to Promptfoo for voice) rather than routing separate components, highlighting advancements in prompt-based voice generation and cloning that reduce the need for distinct STT/LLM/TTS orchestration.