How to make your AI twin (markgreville.ie)

This piece is a practical how‑to for building a personal “AI twin” — a voice‑cloned, phone‑connected conversational agent — using readily available services: ElevenLabs for voice cloning, VAPI as the telephony/agent layer, and OpenAI (via platform API credits) as the LLM. Steps include creating API keys (ElevenLabs and OpenAI), cloning your voice (instant cloning with ~30s of audio or higher‑quality Professional cloning on ElevenLabs’ Creator plan), exporting the voice ID, wiring the keys into VAPI, creating an assistant (select OpenAI as provider and a model such as GPT‑4o‑Cluster), feeding a long system prompt (the author used a 5,000‑word summary of their writing to mimic tone), selecting ElevenLabs’ ElevenTurbo V2.5 as the TTS model, and configuring a speech‑to‑text engine (Deepgram nova‑2 or Google). VAPI can provision a US phone number, attach the assistant and voice, record calls and transcripts (no live listen), and let you test the bot. Costs noted: ElevenLabs starter ~$5/mo (Professional ~$22), and minimal OpenAI platform credits (~$5) to start. Significance and implications: this shows how quickly sophisticated, phone‑based AI personas can be assembled from off‑the‑shelf APIs, democratizing personalized voice agents for outreach, support, or demos. Key technical implications include the role of a dense system prompt (or fine‑tuning-style context) to reproduce “tone,” the need to manage API keys and voice IDs, and tradeoffs between latency/cost across model choices. It also raises obvious privacy, consent and misuse concerns — if you can clone a voice and hand out a number, you should consider authentication, consent and how the twin might be abused.