Simple Vertex Bridge

A simple Vertex AI proxy that automatically refresh tokens for you.

[中文]

Feature

• OpenAI style chat completion API with token attached
• OpenAI style model list API
• Automatically refresh tokens
• Stream output
• Reused h2 connection

Usage

Prerequisites

• Install uv.

Authentication

There are two ways to authenticate:

1. By gcloud CLI
   • Install gcloud CLI.
   • Run gcloud auth application-default login to authenticate.
2. By service account key
   • Create a service account key in the Google Cloud Console, according to this documentation.
   • Download the json key file.
   • Set the GOOGLE_APPLICATION_CREDENTIALS environment variable to the path of the key file.

Launch

• Note: svbridge-config.json will be created in the current directory.

1. You can run uvx simple-vertex-bridge to launch the bridge directly from pypi.
2. Or clone this repo and enter and run uv sync, then activate venv and run python svbridge.py.

Now your API is ready

• Model List: http://localhost:8086/v1/models, v1 can be omitted.
• API Endpoint: http://localhost:8086/v1/chat/completions, v1 can be omitted.
• API Key: Default is anything (if you specify a key, you must use it), will be replaced with Vertex AI token.

CLI Arguments

The bridge uses a configuration file svbridge-config.json located in the current working directory. It's created automatically on the first run if it doesn't exist.

You can override the configuration using command-line arguments when launching the bridge. These arguments will also update the configuration file.

• -p [PORT], --port [PORT]: Port to listen on (default: 8086).
• -b [BIND], --bind [BIND]: Host address to bind to (default: localhost).
• -k [KEY], --key [KEY]: Specify the API key required for authentication. If not set (default), any key will be accepted.
• --auto-refresh/--no-auto-refresh: Enable/disable automatic background refresh of the Vertex token (default: enabled).
• --filter-model-names/--no-filter-model-names: Enable/disable filtering of common model names in the /models endpoint (default: enabled).
• -h, --help: Show help message.

Example:

# Run publicly on port 8848 and set key 'svb-cRztHvmE50'
python svbridge.py -p 8848 -b 0.0.0.0 -k svb-cRztHvmE50

License

The Unlicense.

TBH I dont care what you do with this code, just dont sue me if it breaks something. uwu