Open Source · MIT · Python

Give your companion a voice that whispers.

The open-source framework for building AI companions with a real voice. One config file, no database lock-in.

The Framework

A companion that speaks, remembers, and stays yours.

X/Friend turns a plain text conversation into someone worth talking to. It is the conversation engine behind companion apps, characters, and voice agents. Bring your own UI, your own database, your own monetization. X/Friend handles the rest.

They speak

Twenty-nine SecretVoice voices across twenty-one languages, from breathy whispers to close-mic murmurs.

They remember

Long conversations stay coherent. Durable facts and scene summaries keep a three-hour talk feeling like minute three.

They see

A consistent look from a fixed reference, with custom styles whenever the scene calls for them.

Swap anything

Venice, Grok, or any OpenAI-compatible brain. Your own database, your own image service. One config line.

Multimodal by Default

They talk, they whisper, they show.

One framework, three channels: text, voice, and images. Here is what a conversation looks like.

Luna online

Hey you... I saved a spot for you right here. What are you waiting for?

Good morning. I missed you.

Say my name... slowly. I want to hear exactly how it sounds coming from you.

Thought of you. Send me one back?

That is you. How?

Quickstart

Three lines to a companion.

Scaffold a character, export your API keys, and talk. Your companion remembers you between sessions.

bashterminal

# install the framework
pip install xfriend

# scaffold a character (name, persona, config)
xfriend init luna

# talk. they remember you tomorrow.
xfriend run --config luna/xfriend.yaml

Or embed them in your product

X/Friend is a library first. Keep your own UI, database, and billing; it handles the turns, the memory, and the voice.

Browse the repository

Bring Your Own Everything

A brain, a body, and a voice. Yours to choose.

Every module speaks a standard protocol, so swapping a provider is a config edit, not a rewrite.

Module	Default	Swap with
The brainthe LLM behind their words	Venice AI any of their models	Grok, LongCat, any OpenAI-compatible endpoint
The voicehow they sound, powered by SecretVoice	SecretVoice 29 voices · 21 languages	any other TTS provider, via a compatible endpoint
The bodythe images they create	Wavespeed Qwen img2img	any image service, your own LoRAs