Voice AI routing and agent infrastructure that benchmarks speech stacks by language/use case, exposes one hosted API, and provides an open-source provider-direct Gateway.
Product sources: Y Combinator S26 company/launch page + official product URL
Speko is a voice-AI routing and developer platform that tries to make the speech stack interchangeable. Its hosted Router exposes one API across speech-to-text, LLM and text-to-speech providers, while the public benchmark surface compares models by language, accuracy, latency and cost. The site also describes managed routing and failover so applications can move between providers without rewriting the surrounding agent stack. 1
The developer surface extends beyond the hosted router. Speko documents managed voice agents, typed SDKs, MCP tooling, LiveKit and Pipecat integration paths, and an open-source provider-direct Gateway that can keep BYOK credentials in the customer process. This gives teams multiple integration boundaries rather than forcing all traffic through one managed runtime. 2
Pricing is unusually explicit for an early infrastructure product. Speko lists its own infrastructure at $0.09 per minute across STT, LLM and TTS, while Router pricing adds 5% to the selected provider's published rate. Every account receives $100 in signup credit. Enterprise pricing is custom. The Router and Speko-infrastructure plans are in public preview and carry no SLA during preview, while enterprise SLA terms are contractual. 3
Speko publishes detailed technical and operational information, but most product-quality evidence remains first-party. The benchmark tables expose model-level measurements and costs, while the privacy policy explains that voice sessions can include audio, transcripts, prompts and related request data routed to selected third-party providers. It states that session recordings are generally retained for 30 days unless another configuration or legal requirement applies, and describes controls around credentials, logging and monitoring. 14
Y Combinator lists Speko as an active Summer 2026 company and describes the same voice-model routing product surface. The vendor status page currently reports the API, runtime, routing and related services as operational. These sources support identity and current availability, not independent routing-quality validation. No substantive independent exact-product review, benchmark or eligible numeric user-rating corpus was found. 56
Speko is best suited to developers building production voice agents who need to compare providers across languages, change models without large integration rewrites, or keep a provider-direct deployment option. It is especially relevant to multilingual products and platform teams that want both hosted routing and lower-level SDK or Gateway control. Teams that require a long operating history, a standard preview SLA or independently validated routing performance have less evidence to rely on today. 132
The strongest aspects are breadth and transparency: one routing layer spans STT, LLM and TTS, pricing is public, benchmark data is exposed at model level, and the architecture includes both hosted and open-source provider-direct paths. LiveKit, Pipecat, SDK and MCP surfaces also make the integration story broader than a single endpoint. 132
The main
maturity and evidence independence. Speko is a 2026 startup, its non-enterprise hosted plans are still in public preview without an SLA, and the benchmark/routing claims are measured and published by the vendor itself. Independent exact-product testing and a meaningful external user-rating corpus are not yet available. 35
Products matched by shared tasks, audience, workflow, product type and capabilities.