Hi HN! I’m Bek, founder of Speko, a platform that finds the optimal combination of speech‑to‑text, LLM, and text‑to‑speech models, given your constraints, from all public benchmarked options and explains why.

Typical production voice agents stack three models—STT, an LLM, and TTS—each with dozens of vendors and new releases monthly. Most teams pick a stack once and never recheck, so they run on outdated, costly models. Speko turns this labor into an API: you send your criteria (accuracy, latency, cost or balanced) and the router returns the best model, with headers showing provider, model names and scores. A new session dials the provider straight from memory, eliminating a control‑plane round‑trip; failover only occurs during connection setup.

Customers range from a founder who didn’t know which models to pick to a property‑management AI stuck on legacy STT/TTS, and a medical team needing Spanish‑aware ASR. In every case we identified the best stack from published benchmarks and they now route through Speko.

The gateway is open‑source (MIT) on GitHub, runs as a sidecar, and supports BYOK. Telemetry is opt‑in. Since launch in late June, usage has grown ~25% per week. Learn more at https://speko.ai/