◦ Prompt · Voice

Make Your Voice AI Feel Human — From Laggy Turn-Taking to Smooth, Low-Latency Conversation in One Session

Paste this into Claude Code, Cursor, or Aider and it'll interview your voice-agent codebase about its existing speech-to-text, text-to-speech, and streaming setup BEFORE touching anything — then tune the whole conversation loop tier by tier: measure where the lag actually is, sharpen end-of-turn detection so it stops waiting, stream the model's thinking and the spoken voice so the first word lands fast, keep a long chat from getting slower over time, and teach it to recognize a natural goodbye so it stops taking the last word. Interview first, then improve one layer at a time, each with before-and-after numbers.

Jun 30, 2026voicelatencyspeech-to-texttext-to-speechdeepgramelevenlabsturn-takingconversationtutorial
◦ 2,080 builders waiting

Want the agent behind these prompts?

These prompts came out of building Trillion. The whole thing is going open-source on GitHub. Drop your email and I'll ping you the moment the repo drops.

I'll send a one-time link to confirm your email. Nothing else.

◦ Next prompt
Build Your Own Voice-First AI Agent — From Empty Repo to a Talking, Tool-Using, Always-On Assistant in One Session →

Paste this into Claude Code, Cursor, or Codex and it'll interview you, help you name your agent, then walk your codebase through building a voice-first assistant tier by tier: a text conversation loop you can debug before you ever add audio, a tool registry that lets the agent actually do things, real speech-in and speech-out (Deepgram for transcription, ElevenLabs for voice) so you talk to it instead of typing, memory that survives a restart, an always-on background loop so the agent can reach out to you first, and the safety rails that keep a proactive assistant from doing something you didn't ask for. Interview first, then build tier by tier. Each tier runs on its own and is verified before the next begins.