RealChar. - Your Realtime AI Character

🎙️🤖Create, customize and talk to your AI Character/Companion in realtime🎙️🤖

✨ Demo

Try our site at RealChar.ai

Not sure how to pronounce RealChar? Listen to this 👉 audip

Demo 1 - with Santa Claus!

Santa_tw.mp4

Demo 2 - with AI Elon about cage fight!

elon-edit-camera.mp4

Demo 3 - with AI Raiden about AI and "real" memory

raiden.mp4

Demo settings: Web, GPT4, ElevenLabs with voice clone, Chroma, Google Speech to Text

🎯 Key Features

• Easy to use: No coding required to create your own AI character.
• Customizable: You can customize your AI character's personality, background, and even voice
• Realtime: Talk to or message your AI character in realtime
• Multi-Platform: You can talk to your AI character on web, terminal and mobile(Yes. we open source our mobile app)
• Most up-to-date AI: We use the most up-to-date AI technology to power your AI character, including OpenAI, Anthropic Claude 2, Chroma, Whisper, ElevenLabs, etc.
• Modular: You can easily swap out different modules to customize your flow. Less opinionated, more flexible. Great project to start your AI Engineering journey.

🔬 Tech stack

• ✅Web: React JS, Vanilla JS, WebSockets
• ✅Mobile: Swift, WebSockets
• ✅Backend: FastAPI, SQLite, Docker
• ✅Data Ingestion: LlamaIndex, Chroma
• ✅LLM Orchestration: LangChain, Chroma
• ✅LLM: ReByte, OpenAI GPT3.5/4, Anthropic Claude 2, Anyscale Llama2
• ✅Speech to Text: Local WhisperX, Local Whisper, OpenAI Whisper API, Google Speech to Text
• ✅Text to Speech: ElevenLabs, Edge TTS, Google Text to Speech
• ✅Voice Clone: ElevenLabs

📚 Comparison with existing products

👨‍🚀 Prerequisites

Before setting up this project, follow these instructions to meet the prerequisites:

1. Large Language Models

1.1 ReByte API Key

To get your ReByte API key, follow these steps:

1. Go to the ReByte website and sign up for an account if you haven't already.
2. Once you're logged in, go to Settings > API Keys.
3. Generate a new API key by clicking on the "Generate" button.
4. Prepare to use it during configuration.

1.2 (Optional) OpenAI API Token

👇click me

This application utilizes the OpenAI API to access its powerful language model capabilities. In order to use the OpenAI API, you will need to obtain an API token.

To get your OpenAI API token, follow these steps:

1. Go to the OpenAI website and sign up for an account if you haven't already.
2. Once you're logged in, navigate to the API keys page.
3. Generate a new API key by clicking on the "Create API Key" button.
4. Prepare to use it during configuration.

(Optional) To use Azure OpenAI API instead, refer to the following section:

1. Set API type in your .env file: OPENAI_API_TYPE=azure

If you want to use the earlier version 2023-03-15-preview:

OPENAI_API_VERSION=2023-03-15-preview

2. To set the base URL for your Azure OpenAI resource. You can find this in the Azure portal under your Azure OpenAI resource.

OPENAI_API_BASE=https://your-base-url.openai.azure.com

3. To set the OpenAI model deployment name for your Azure OpenAI resource.

OPENAI_API_MODEL_DEPLOYMENT_NAME=gpt-35-turbo-16k

4. To set the OpenAIEmbeddings model deployment name for your Azure OpenAI resource.

OPENAI_API_EMBEDDING_DEPLOYMENT_NAME=text-embedding-ada-002

1.3 (Optional) Anthropic(Claude 2) API Token

👇click me

To get your Anthropic API token, follow these steps:

1. Go to the Anthropic website and sign up for an account if you haven't already.
2. Once you're logged in, navigate to the API keys page.
3. Generate a new API key by clicking on the "Create Key" button.
4. Prepare to use it during configuration.

1.4 (Optional) Anyscale API Token

👇click me

To get your Anyscale API token, follow these steps:

1. Go to the Anyscale website and sign up for an account if you haven't already.
2. Once you're logged in, navigate to the Credentials page.
3. Generate a new API key by clicking on the "Generate credential" button.
4. Prepare to use it during configuration.

2. Speech to Text

We support whisperX as the local speech to text engine. Work with CPU and NVIDIA GPU.

2.1 (Optional) Google Speech-to-Text API

👇click me

To get your Google Cloud API credentials.json, follow these steps:

1. Go to the GCP website and sign up for an account if you haven't already.
2. Follow the guide to create a project and enable Speech to Text API
3. Put google_credentials.json in the root folder of this project. Check GCP website
4. Change SPEECH_TO_TEXT_USE to use GOOGLE in your .env file

2.2 (Optional) OpenAI Whisper API

👇click me

3. Text to Speech

Edge TTS is the default and is free to use.

3.1 (Optional) ElevenLabs API Key

👇click me

1. Creating an ElevenLabs Account

   Visit ElevenLabs to create an account. You'll need this to access the text to speech and voice cloning features.

2. In your Profile Setting, you can get an API Key.

3. Prepare to use it during configuration.

3.2 (Optional) Google Text-to-Speech API

💿 Installation via Python

• Step 1. Clone the repo

  git clone https://github.com/Shaunwei/RealChar.git && cd RealChar

• Step 2. Install requirements

  Install portaudio and ffmpeg for audio

  # for mac
  brew install portaudio
  brew install ffmpeg

  # for ubuntu
  sudo apt update
  sudo apt install portaudio19-dev
  sudo apt install ffmpeg

  Note:

  • ffmpeg>=4.4 is needed to work with torchaudio>=2.1.0

  • Mac users may need to add ffmpeg library path to DYLD_LIBRARY_PATH for torchaudio to work:

    export DYLD_LIBRARY_PATH=/opt/homebrew/lib:$DYLD_LIBRARY_PATH

  Then install all python requirements

  pip install -r requirements.txt

  If you need a faster local speech to text, install whisperX

  pip install git+https://github.com/m-bain/whisperx.git

• Step 3. Create an empty sqlite database if you have not done so before

  sqlite3 test.db "VACUUM;"

• Step 4. Run db upgrade

  alembic upgrade head

  This ensures your database schema is up to date. Please run this after every time you pull the main branch.

• Step 5. Setup .env:

  cp .env.example .env

  Update API keys and configs following the instructions in the .env file.

  Note that some features require a working login system. You can get your own OAuth2 login for free with Firebase if needed. To enable, set USE_AUTH to true and fill in the FIREBASE_CONFIG_PATH field. Also fill in Firebase configs in client/next-web/.env.

• Step 6. Run backend server with cli.py or use uvicorn directly

  python cli.py run-uvicorn
  # or
  uvicorn realtime_ai_character.main:app

• Step 7. Run frontend client:

  • web client:

    Create an .env file under client/next-web/

    cp client/next-web/.env.example client/next-web/.env

    Adjust .env according to the instruction in client/next-web/README.md.

    Start the frontend server:

    python cli.py next-web-dev
    # or
    cd client/next-web && npm run dev
    # or
    cd client/next-web && npm run build && npm run start

    After running these commands, a local development server will start, and your default web browser will open a new tab/window pointing to this server (usually http://localhost:3000).

  • (Optional) Terminal client:

    Run the following command in your terminal

    python client/cli.py

  • (Optional) mobile client:

    open client/mobile/ios/rac/rac.xcodeproj/project.pbxproj in Xcode and run the app

• Step 8. Select one character to talk to, then start talking. Use GPT4 for better conversation and Wear headphone for best audio(avoid echo)

Note if you want to remotely connect to a RealChar server, SSL set up is required to establish the audio connection.

(Optional) 📀 Installation via Docker

👇click me

1. Use the docker-compose.yaml to run the app. Run the following command after configuring your .env file in the root directory:

   docker-compose up

   If you have issues with docker (especially on a non-Linux machine), please refer to https://docs.docker.com/get-docker/ (installation) and https://docs.docker.com/desktop/troubleshoot/overview/ (troubleshooting).

2. Go to http://localhost:3000 to start talking or use terminal client with

   python client/cli.py

(Optional) 🔥 Create Your Own Characters

👇click me

Create Characters Locally

see realtime_ai_character/character_catalog/README.md

Create Characters on ReByte.ai

see docs/rebyte_agent_clone_instructions.md

(Optional) ☎️ Twilio Integration

👇click me

To use Twilio with RealChar, you need to set up a Twilio account. Then, fill in the following environment variables in your .env file:

TWILIO_ACCOUNT_SID=YOUR_TWILIO_ACCOUNT_SID
TWILIO_ACCESS_TOKEN=YOUR_TWILIO_ACCESS_TOKEN
DEFAULT_CALLOUT_NUMBER=YOUR_PHONE_NUMBER

You'll also need to install torch and torchaudio to use Twilio.

Now, you can receive phone calls from your characters by typing /call YOURNUMBER in the text box when chatting with your character.

Note: only US phone numbers and Elevenlabs voiced characters are supported at the moment.

🆕! Anyscale and LangSmith integration

👇click me

Anyscale

You can now use Anyscale Endpoint to serve Llama-2 models in your RealChar easily! Simply register an account with Anyscale Endpoint. Once you get the API key, set this environment variable in your .env file:

ANYSCALE_ENDPOINT_API_KEY=<your API Key>

By default, we show the largest servable Llama-2 model (70B) in the Web UI. You can change the model name (meta-llama/Llama-2-70b-chat-hf) to other models, e.g. 13b or 7b versions.

LangSmith

If you have access to LangSmith, you can edit these environment variables to enable:

LANGCHAIN_TRACING_V2=false # default off
LANGCHAIN_ENDPOINT=https://api.smith.langchain.com
LANGCHAIN_API_KEY=YOUR_LANGCHAIN_API_KEY
LANGCHAIN_PROJECT=YOUR_LANGCHAIN_PROJECT

And it should work out of the box.

📍 Roadmap

• Launch v0.0.4
• Create a new character via web UI
• Lower conversation latency
• Support Twilio
• Support ReByte
• Persistent conversation*
• Session management*
• Support RAG*
• Support Agents/GPTs*
• Add additional TTS service*

$*$ These features are powered by ReByte platform.

🫶 Contribute to RealChar

Please check out our Contribution Guide!

💪 Contributors

🎲 Community

• Join us on Discord