Speko (YC S26) – OpenRouter for Voice AI
A multi-model evaluation and routing layer for voice agent development teams, selecting STT, LLM, and TTS combinations based on language, region, and cost/delay/accuracy targets.
Who it is for: Teams needing to continuously select and replace STT, LLM, and TTS in multilingual production voice agents; developers using LiveKit, Pipecat, or self-built voice agent infrastructure compatible with OpenAI API; teams wanting to retain their own infrastructure but reduce the cost of model evaluation and vendor switching.
Core capabilities
- Routing by language and target: Filter and select models based on language, region, accuracy, delay, cost, or balance targets.
- Multi-model benchmark evaluation: Evaluate multiple speech models by language and date, and publish evaluation boards.
- Unified gateway access: Connect multiple vendors through a unified base URL, key, and OpenAI API-compatible interface.
- Session prefetch and connection failure switching: Prefetch signed session plans and attempt candidate vendors if connection setup fails.
Pricing
Specific pricing, billing units, free quotas, and margin structure are not disclosed.
Updated 2026-08-18