Speko (YC S26): The "OpenRouter" for Voice AI

Speko is positioning itself as the definitive router for voice models, providing a unified interface to access a multitude of speech technologies. Backed by Y Combinator, Speko streamlines the process of integrating voice AI by wiring every major speech model into a single, cohesive API.

The Core Value Proposition: Instead of relying on vendor-supplied English leaderboards, Speko uses published measurements to determine the optimal model based on the specific language and objective of the request.

🛠️ How the Routing Works

Speko acts as an intelligent intermediary. Rather than hard-coding a specific provider, the system dynamically selects the best tool for the job.

📊 Benchmarking & Language Coverage

A critical insight from Speko's data is the disparity in language support. While many models claim global readiness, the actual data tells a different story:

• The English Bias: $\approx 47.8\%$ (11 of 23) of measured models are only benchmarked in English.
• The "Top Model" Fallacy: The model ranking #1 in English may have an unknown rank in other languages.
• Diversified Winners: Across 9 different languages, 4 different models emerged as the winners.

Speech-to-Text (STT) Performance Matrix

The following table outlines the trade-off between accuracy (measured by Word Error Rate) and cost.

Note: Some models like deepgram:flux-general-en and openai:gpt-live-transcribe are available but lack specific WER data in this view.

Understanding the Metric

The primary metric used for accuracy is the Word Error Rate (WER), calculated as: $WER = \frac{S + D + I}{N}$ (Where $S$ = substitutions, $D$ = deletions, $I$ = insertions, and $N$ = number of words in the reference).

🔌 Integration & Gateway

Speko provides a single base URL and one API key that works across all providers. Because it is OpenAI API compatible, it integrates seamlessly into existing frameworks.

Implementation Example (LiveKit Agents)

import { defineAgent, voice } from '@livekit/agents';
import * as openai from '@livekit/agents-plugin-openai';

const key = process.env.SPEKO_API_KEY!;
const baseURL = 'https://api.speko.ai/v1';

export default defineAgent({
  entry: async (ctx) => {
    const session = new voice.AgentSession({
      stt: new openai.STT({ apiKey: key, baseURL, model: 'auto' }),
      llm: new openai.LLM({ apiKey: key, baseURL, model: 'auto' }),
      tts: new openai.TTS({ apiKey: key, baseURL, model: 'auto' }),
    });
    await session.start({ agent: new voice.Agent({ ... }) });
  },
});

MCP Support

You can also point your agent at Speko via the Model Context Protocol (MCP):

curl command for Claude MCP:

claude mcp add --transport http speko https://mcp.speko.ai/mcp

🚀 Getting Started Checklist

• Create an account and sign in.
• Generate your SPEKO_API_KEY.
• Compare models using the benchmark dashboard.
• Update your baseURL to https://api.speko.ai/v1.
• (Optional) Book a demo with the founder.