One API on Workers

A distributed, low-latency, high-performance AI gateway on Cloudflare Workers, with channel management, load balancing, and usage analytics.

English | 中文

Based on dreamhunter2333/awsl-one-api, further developed with AI assistance.

Overview

• Built on Cloudflare Workers: global edge network, low latency, high performance
• Unified proxy entry: /v1/chat/completions, /v1/messages, /v1/responses, /v1/audio/speech, /v1/models
• Weighted load balancing: multi-key per channel, retry, key rotation, and cross-channel fallback
• Quotas and billing: token-level quotas, global model pricing, channel-level model pricing
• Observability: writes to Cloudflare Analytics Engine; admin UI covers overview, trends, distributions, and usage log search
• Admin console: React + Vite UI for channels, tokens, pricing, API testing, and system settings
• Admin security: admin-token login by default; optional Telegram 2FA; rate limits and session cookies on the login path
• API docs: Chanfana-powered Swagger, ReDoc, and OpenAPI JSON, toggleable in system settings

Supported Routes

Proxy APIs

Route	Description	Channel types
/v1/chat/completions	OpenAI Chat Completions–compatible proxy	openai, azure-openai, gemini
/v1/messages	Anthropic Claude Messages–compatible proxy	claude, claude-to-openai
/v1/responses	OpenAI / Azure Responses proxy	openai-responses, azure-openai-responses
/v1/audio/speech	TTS speech generation proxy	openai-audio, azure-openai-audio
/v1/models	Model list filtered by token permissions and quota	Returned dynamically from channels the token can access

Deployment

One-click deploy

Manual deploy

# Create a D1 database and note its name and ID
# Enable Analytics Engine, create a dataset, and note its name
# Put those values into wrangler.jsonc

# Set production secrets
wrangler secret put ADMIN_TOKEN
wrangler secret put CF_API_TOKEN
wrangler secret put CF_ACCOUNT_ID

# Publish the Worker
bun run deploy

Admin console

Current admin pages:

• Dashboard: total requests, success rate, cost, latency, and token/channel/model/provider distributions
• Usage Logs: filter detailed logs by time range, dimensions, keywords, and result
• Channels: channel config, weights, auto retry/rotation, model mapping, pull upstream model lists
• Tokens: API token management, channel access scope, quotas, usage reset
• Pricing: global model pricing; per-token and per-request billing
• API Test: test /v1/chat/completions, /v1/messages, /v1/responses, /v1/audio/speech from the console
• System Settings: Telegram admin verification, currency display precision, API docs toggle

Project Structure

one-api-workers/
├── src/
│   ├── admin/                    # Admin APIs: auth / channel / token / pricing / analytics / system
│   ├── analytics/                # Analytics Engine write and query
│   ├── db/                       # D1 init and migrations
│   ├── providers/                # Upstream provider proxies
│   ├── billing.ts                # Billing and amount precision
│   ├── channel-config.ts         # Channel config normalization
│   ├── system-config.ts          # System config and Telegram security settings
│   └── index.ts                  # Worker entry
├── frontend/
│   ├── src/pages/                # Dashboard / Channels / Tokens / Pricing / Usage Logs / Settings
│   ├── src/components/           # Layout, charts, UI components
│   └── package.json              # Frontend build and lint
├── public/                       # Frontend build output, served by the Worker
├── docs/                         # Usage and security docs
├── tests/                        # Mock upstream + local E2E scripts
├── wrangler.jsonc                # Production config
├── wrangler.local.jsonc          # Local Worker config
├── type.d.ts                     # Worker bindings and shared types
└── package.json                  # Root workspace and dev scripts

Quick Start

Requirements

• Bun 1.3+
• Cloudflare account with Workers + D1 + Analytics Engine (usage_events_by_token)

Install dependencies

bun install

Configure Cloudflare bindings

wrangler.jsonc / wrangler.local.jsonc already include the required binding shape; replace them with your own values:

• d1_databases[].database_name / database_id: your D1 database
• analytics_engine_datasets[].dataset: defaults to usage_events_by_token
• vars.FRONTEND_DEV_SERVER_URL: local integrated dev only; default http://127.0.0.1:5173
• assets: keep the public/ → ASSETS binding

Required secrets in the current setup:

• ADMIN_TOKEN: admin login token (required)
• CF_API_TOKEN: Analytics Engine SQL queries for the dashboard and usage logs
• CF_ACCOUNT_ID: used with CF_API_TOKEN for Cloudflare Analytics queries

Example:

wrangler secret put ADMIN_TOKEN
wrangler secret put CF_API_TOKEN
wrangler secret put CF_ACCOUNT_ID

For local development, put these in .dev.vars; scripts under tests/ also prefer that file.

Database init and migrations

D1 schema init and migrations run automatically on the first request. No manual SQL is required.

To trigger init after deploy (with admin auth):

POST /api/admin/db_initialize

Local development

Start frontend + Worker together:

bun run dev

Common URLs:

• Vite frontend: http://127.0.0.1:5173
• Local Worker: http://127.0.0.1:8787

Other useful commands:

bun run dev:worker
bun run build
bun run cf-typegen
cd frontend && bun run lint

Documentation

• Usage and configuration (中文)
• Admin auth and protection (中文)
• Testing and verification (中文)

Contributing

Issues and pull requests are welcome.

License

MIT License

Credits: dreamhunter2333/awsl-one-api

Support

For questions or suggestions, open an Issue or contact the maintainers.