One API

The One API frontend supports multiple themes. Note that not all themes are guaranteed to be synchronized with all features immediately; the default theme is prioritized for updates.

• default: The primary theme developed by [JustSong].
• berry: A theme developed by [MartialBE].
• air: A theme developed by [Calon].

For specific development instructions regarding the berry theme, refer to web/berry/README.md.

For high availability and scaling across multiple servers, follow these requirements:

1. Shared Session: Set the same SESSION_SECRET on all servers.
2. Shared Database: Use MySQL instead of SQLite via SQL_DSN. All nodes must connect to the same database.
3. Node Roles: Set NODE_TYPE=slave on all nodes except the master.
4. Configuration Sync: Set SYNC_FREQUENCY (in seconds) to periodically sync configurations from the database.
5. Frontend Redirection: On slave nodes, optionally set FRONTEND_BASE_URL to redirect page requests to the master server.
6. Caching: Install Redis on slave nodes and configure REDIS_CONN_STRING to achieve zero-latency access to cached data.
7. Master Caching: If the master node experiences high database latency, enable Redis and set SYNC_FREQUENCY.

One API calculates quota based on a multiplier system to account for different model costs.

Formula: Quota = Group multiplier * Model multiplier * (number of prompt tokens + number of completion tokens * completion multiplier)

Key Details:

• Completion Multiplier: Fixed at 1.33 for GPT-3.5 and 2 for GPT-4, matching official definitions.
• Token Counts: For non-stream modes, the official API returns total tokens. Note that prompt and completion multipliers are applied separately.

By default, One API uses load balancing to distribute requests across available channels. To force a request to use a specific channel, append the CHANNEL_ID to your token using a hyphen.

Requirement: The token must be created by an administrator to support channel ID specification.

Format: Authorization: Bearer ONE_API_KEY-CHANNEL_ID

One API uses JSON for all requests and responses. A standard successful response body follows this structure:

{
  "message": "请求信息",
  "success": true,
  "data": {}
}

One API calculates quotas using a specific formula involving group multipliers, model multipliers, and token counts. Note that completion tokens may have a different multiplier than prompt tokens.

Quota Formula: Quota = Group Multiplier * Model Multiplier * (Prompt Tokens + Completion Tokens * Completion Multiplier)

Fixed Completion Multipliers:

• GPT-3.5: 1.33
• GPT-4: 2.0

Note: In non-streaming mode, the official API returns the total tokens consumed, but you must account for the different multipliers for prompt and completion tokens.

If you prefer Docker Compose, you can use the following setup. Note that the parameter settings remain the same as the standard Docker deployment.

# Start the service (supports MySQL startup)
docker-compose up -d

# Check deployment status
docker-compose ps

To deploy manually, you must build both the frontend and the backend.

1. Clone the repository:

   git clone https://github.com/songquanpeng/one-api.git

2. Build the Frontend:

   cd one-api/web/default
   npm install
   npm run build

3. Build the Backend:

   cd ../..
   go mod download
   go build -ldflags "-s -w" -o one-api

4. Run the application:

   chmod u+x one-api
   ./one-api --port 3000 --log-dir ./logs

Default Credentials:

• Username: root
• Password: 123456