TransFire is a simple tool that allows you to use your locally running LLMs while far from home, without requiring port forwarding. TransFire will route an OpenAI compatible API exposed by LMStudio or Ollama through your firebase instance of choice, encrypting all traffic with a pre-shared AES key so not even Google will be able to read your conversations.

To get started download and install the apk and follow the setup instructions below.

First you will need to setup the client:

1. Install the APK on your phone
2. Go to the Firebase console.
3. Click on Create a new Firebase project
4. Proceed through the whole process, opting out of Analytics and Gemini preferably
5. Now on the left panel, expand the Build dropdown and select Realtime Database
6. Click on Create Database
7. Select a database region of your choice
8. Select Start in locked mode and proceed
9. Click on the URL icon to copy the database URL. This is your Firebase Database URL
10. Now click on the settings icon in the top-left corner and open Project Settings
11. Go to the Service accounts tab and then click on Database secrets
12. You should see one secret in the list, if not click on Add secret
13. Hover on the secret to reveal the Show button and click it, then copy the key. This is your Firebase Database API key
14. Now go to the app and click on Get started
15. Put the Firebase Database URL and Firebase Database API key in the corresponding fields, then choose an AES password to encrypt the traffic to/from Firebase and put it into Encryption password.
16. Click on Save configuration and then Next.
17. You can now proceed to server configuration

Now you can setup and start the server:

1. Clone the repository with git clone https://github.com/Belluxx/TransFire
2. Navigate to the server directory with cd TransFire/transfire-server/
3. Create a virtual environment with python3 -m venv .venv
4. Activate it with source .venv/bin/activate
5. Install dependencies with pip3 install -r requirements.txt
6. Now copy example.env to .env
7. Open the new .env file
8. Fill in the fields FIREBASE_URL, FIREBASE_API_KEY, ENCRYPTION_PASSWORD with the same values used during the client setup
9. Choose a POLL_INTERVAL, not too low (use >2 seconds) or you risk finishing your free firebase daily usage
10. Put the correct OPENAI_LIKE_API_URL. It will be http://127.0.0.1:1234 if you are using LMStudio or http://127.0.0.1:11434 if you are using Ollama
11. OPENAI_LIKE_API_KEY should stay as is, change it only if you know what you are doing (for example using remote APIs)
12. Start the server with python3 server.py

sequenceDiagram
    participant TransFire App
    participant Firebase
    participant TransFire Server
    participant LLM Server

    TransFire App->>TransFire App: Append user message to chat
    TransFire App->>TransFire App: Encrypt chat
    TransFire App->>Firebase: Send chat
    
    TransFire Server->>Firebase: Get chat
    TransFire Server->>TransFire Server: Decrypt chat
    TransFire Server->>LLM Server: Send chat
    LLM Server->>TransFire Server: Return LLM response
    TransFire Server->>TransFire Server: Encrypt response
    TransFire Server->>Firebase: Send response
    
    TransFire App->>Firebase: Get response
    TransFire App->>TransFire App: Decrypt response
    TransFire App->>TransFire App: Append response to chat

• No support for two simultaneous clients using the same Firebase Database
• No automatic detection of available models due to the heterogeneity of Ollama, LMStudio, etc...
• No support for chat history and multiple chats (will be added in the future)

Thanks to compose-richtext I was able to add markdown parsing that is essential to make LLM output readable.