A ChatGPT client that actually talks back

What it does

Chat with GPT is a React/TypeScript web interface for the OpenAI ChatGPT API. You supply your own API key, chat in the browser, and optionally have replies read aloud via ElevenLabs text-to-speech or your browser’s built-in voice. It keeps conversation history locally and lets you search past chats, tweak the hidden system prompt, and adjust response temperature.

The interesting bit

The project treats voice as a first-class citizen rather than a gimmick: it wires in both high-fidelity ElevenLabs synthesis and OpenAI Whisper speech recognition, turning a text API into a hands-free intercom. That, plus exposing the system prompt and temperature controls, makes it feel like a power-user skin for ChatGPT.

Key highlights

• Realistic voice output via ElevenLabs, with browser TTS fallback and OpenAI Whisper speech input.
• Searchable chat history, editable system prompts, and a temperature slider.
• Public share URLs for specific conversations.
• Full markdown support for code, tables, and math.
• API keys stay client-side by default; self-hosted instances can optionally store them in a server-side config.yaml.

Caveats

• The README explicitly warns that storing API keys server-side exposes them to anyone with access to your self-hosted instance.

Verdict

Grab this if you want a more configurable, voice-enabled alternative to the official ChatGPT UI and are comfortable bringing your own API keys. Pass if you need local LLM inference or a managed multi-user backend.