pleasefixai/speko

Use Speko to transcribe audio, synthesize speech, and pick the model for each leg of a voice pipeline from measured benchmarks instead of a hardcoded vendor. One key covers STT, LLM and TTS across 50+ models, with a dry-run route preview, per-language selection, price ceilings and automatic failover. Use when asked to transcribe a recording or voice note, read something aloud, choose or justify a speech model, work in a language other than English, or cap voice spend. For placing phone calls, use the speko-calls skill instead.

Place and monitor outbound AI phone calls through Speko. This skill dials real telephone numbers and costs real money, so it reads its own environment variable and asks the user to confirm the number and the purpose before every call. Use when explicitly asked to call someone, ring a number, run an outbound voice campaign, or check the status, recording or transcript of a call that was already placed. For transcription, speech synthesis or model routing with no dialing involved, use the speko skill instead.