aiclient-api

English | 中文

A unified AI gateway daemon that authenticates against GitHub Copilot and Kiro (AWS CodeWhisperer), exposing standard OpenAI-compatible and Anthropic-compatible API endpoints locally.

Use your existing Copilot or Kiro subscription with any tool that speaks the OpenAI or Anthropic API — Claude Code, Cursor, Continue, or your own scripts.

Features

• Dual provider support — GitHub Copilot (Individual / Business / Enterprise) and Kiro (Builder ID / Google / GitHub / IAM Identity Center)
• OpenAI-compatible endpoint — POST /v1/chat/completions, GET /v1/models
• Anthropic-compatible endpoint — POST /v1/messages
• Usage tracking — GET /v1/usage to monitor token consumption by provider and model
• Automatic format conversion — OpenAI <-> Anthropic, transparently
• Passthrough mode — Skip conversion when the provider natively supports the target format
• SSE streaming — Full streaming support with format-aware chunk conversion
• Smart model routing — By model prefix (copilot/gpt-4), X-Provider header, or config default
• Bearer token auth — Optional API key protection on all /v1/* routes
• Per-IP rate limiting — Configurable request interval
• Daemon mode — Runs in background with PID management, or --foreground for debugging
• Unix socket control — CLI commands talk to the running daemon over a Unix socket
• Hot config reload — SIGHUP or config reload to apply changes without restart
• XDG-compliant paths — Config, tokens, logs, PID, socket all follow XDG conventions

Quick Start

1. Build

cargo build --release

2. Authenticate

# GitHub Copilot (device flow — opens browser)
aiclient-api auth copilot

# Kiro / AWS Builder ID (interactive menu)
aiclient-api auth kiro

# Kiro / IAM Identity Center (organization identity)
aiclient-api auth kiro --start-url https://my-org.awsapps.com/start --region us-east-1

3. Start the daemon

# Background mode (default)
aiclient-api start

# Foreground mode (for debugging)
aiclient-api start --foreground

# Custom port and API key
aiclient-api start --port 8080 --api-key my-secret

4. Use it

# OpenAI format
curl http://127.0.0.1:9090/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "copilot/gpt-4o",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

# Anthropic format
curl http://127.0.0.1:9090/v1/messages \
  -H "Content-Type: application/json" \
  -d '{
    "model": "copilot/claude-sonnet-4",
    "messages": [{"role": "user", "content": "Hello!"}],
    "max_tokens": 1024
  }'

# Kiro provider
curl http://127.0.0.1:9090/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "kiro/claude-sonnet-4-6",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

# List available models
curl http://127.0.0.1:9090/v1/models

# Check usage statistics
curl http://127.0.0.1:9090/v1/usage

Usage Tracking

Monitor your token consumption in real-time:

# Get usage statistics
curl http://127.0.0.1:9090/v1/usage | jq .

# Example response
{
  "providers": {
    "copilot": {
      "request_count": 10,
      "input_tokens": 150,
      "output_tokens": 200,
      "total_tokens": 350,
      "models": {
        "gpt-4o": { "request_count": 5, "input_tokens": 80, ... },
        "claude-sonnet-4.5": { "request_count": 5, "input_tokens": 70, ... }
      }
    },
    "kiro": { ... }
  },
  "total": {
    "request_count": 10,
    "input_tokens": 150,
    "output_tokens": 200,
    "total_tokens": 350
  }
}

# Reset statistics
curl -X DELETE http://127.0.0.1:9090/v1/usage

Features:

• ✅ Real-time tracking per provider and model
• ✅ Tracks input/output tokens separately
• ✅ Aggregated total statistics
• ⚠️ In-memory storage (resets on restart)

See USAGE_TRACKING.md for full documentation.

Configuration

Config file: ~/.config/aiclient-api/config.toml

default_format = "openai"       # "openai" or "anthropic"
default_provider = "copilot"    # default provider when no prefix
api_key = ""                    # empty = no auth required

[server]
host = "127.0.0.1"
port = 9090
rate_limit_seconds = 0          # 0 = disabled

[providers.copilot]
type = "copilot"
enabled = true
account_type = "individual"     # "individual", "business", "enterprise"

[providers.kiro]
type = "kiro"
enabled = true
region = "us-east-1"

[logging]
level = "info"

CLI Commands

Model Routing

Models are routed by three mechanisms, in priority order:

1. Model prefix — copilot/gpt-4o routes to the copilot provider with model gpt-4o
2. X-Provider header — X-Provider: kiro routes to the kiro provider
3. Config default — Falls back to default_provider in config

API Endpoints

Architecture

src/
  auth/           # Token management
    copilot.rs    # GitHub device flow OAuth
    kiro.rs       # Builder ID + social auth (PKCE)
    token_store.rs
  cli/            # CLI commands
  config/         # TOML config loading
  convert/        # OpenAI <-> Anthropic format conversion
    stream.rs     # SSE chunk conversion
  daemon/         # Process management + Unix socket control
  providers/      # Provider implementations
    copilot/      # GitHub Copilot (VSCode header spoofing)
    kiro/         # Kiro / CodeWhisperer API
    router.rs     # Model-based provider routing
  routes/         # HTTP route handlers
    usage.rs      # Usage tracking endpoints
  server/         # Axum server, middleware (auth, rate-limit, CORS)
  usage/          # Token usage tracking
    tracker.rs    # UsageTracker implementation
  util/           # XDG paths, error types, streaming helpers

File Locations

License

MIT

aiclient-api

English | 中文

统一 AI 网关守护进程，通过认证接入 GitHub Copilot 和 Kiro（AWS CodeWhisperer），在本地暴露标准的 OpenAI 兼容和 Anthropic 兼容 API 端点。

用你已有的 Copilot 或 Kiro 订阅，配合任何支持 OpenAI 或 Anthropic API 的工具使用 — Claude Code、Cursor、Continue，或你自己的脚本。

功能特性

• 双 Provider 支持 — GitHub Copilot（Individual / Business / Enterprise）和 Kiro（Builder ID / Google / GitHub / IAM Identity Center）
• OpenAI 兼容端点 — POST /v1/chat/completions、GET /v1/models
• Anthropic 兼容端点 — POST /v1/messages
• 使用量跟踪 — GET /v1/usage 实时监控按 Provider 和模型的 Token 消耗
• 自动格式转换 — OpenAI <-> Anthropic 双向透明转换
• 直通模式 — Provider 原生支持目标格式时跳过转换，直接透传
• SSE 流式 — 完整的流式支持，带格式感知的 chunk 转换
• 智能模型路由 — 通过模型前缀（copilot/gpt-4）、X-Provider 请求头或配置默认值
• Bearer Token 认证 — 可选的 API Key 保护，作用于所有 /v1/* 路由
• IP 限流 — 可配置的请求频率限制
• 守护进程模式 — 后台运行 + PID 管理，或 --foreground 前台调试
• Unix Socket 控制 — CLI 命令通过 Unix Socket 与运行中的守护进程通信
• 热重载配置 — SIGHUP 信号或 config reload 命令即时生效
• XDG 路径规范 — 配置、Token、日志、PID、Socket 均遵循 XDG 目录规范

快速开始

1. 构建

cargo build --release

2. 认证

# GitHub Copilot（设备码流程 — 自动打开浏览器）
aiclient-api auth copilot

# Kiro / AWS Builder ID（交互式菜单）
aiclient-api auth kiro

# Kiro / IAM Identity Center（组织身份认证）
aiclient-api auth kiro --start-url https://my-org.awsapps.com/start --region us-east-1

3. 启动守护进程

# 后台模式（默认）
aiclient-api start

# 前台模式（调试用）
aiclient-api start --foreground

# 自定义端口和 API Key
aiclient-api start --port 8080 --api-key my-secret

4. 使用

# OpenAI 格式
curl http://127.0.0.1:9090/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "copilot/gpt-4o",
    "messages": [{"role": "user", "content": "你好！"}]
  }'

# Anthropic 格式
curl http://127.0.0.1:9090/v1/messages \
  -H "Content-Type: application/json" \
  -d '{
    "model": "copilot/claude-sonnet-4",
    "messages": [{"role": "user", "content": "你好！"}],
    "max_tokens": 1024
  }'

# 使用 Kiro Provider
curl http://127.0.0.1:9090/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "kiro/claude-sonnet-4-6",
    "messages": [{"role": "user", "content": "你好！"}]
  }'

# 列出可用模型
curl http://127.0.0.1:9090/v1/models

# 查询使用统计
curl http://127.0.0.1:9090/v1/usage

使用量跟踪

实时监控你的 Token 消耗：

# 获取使用统计
curl http://127.0.0.1:9090/v1/usage | jq .

# 响应示例
{
  "providers": {
    "copilot": {
      "request_count": 10,
      "input_tokens": 150,
      "output_tokens": 200,
      "total_tokens": 350,
      "models": {
        "gpt-4o": { "request_count": 5, "input_tokens": 80, ... },
        "claude-sonnet-4.5": { "request_count": 5, "input_tokens": 70, ... }
      }
    },
    "kiro": { ... }
  },
  "total": {
    "request_count": 10,
    "input_tokens": 150,
    "output_tokens": 200,
    "total_tokens": 350
  }
}

# 重置统计数据
curl -X DELETE http://127.0.0.1:9090/v1/usage

功能特点：

• ✅ 按 Provider 和模型实时跟踪
• ✅ 分别统计输入/输出 tokens
• ✅ 聚合总计统计
• ⚠️ 内存存储（重启后清零）

详细文档参见 USAGE_TRACKING.md。

配置

配置文件路径：~/.config/aiclient-api/config.toml

default_format = "openai"       # "openai" 或 "anthropic"
default_provider = "copilot"    # 无前缀时的默认 Provider
api_key = ""                    # 留空 = 不需要认证

[server]
host = "127.0.0.1"
port = 9090
rate_limit_seconds = 0          # 0 = 不限流

[providers.copilot]
type = "copilot"
enabled = true
account_type = "individual"     # "individual"、"business"、"enterprise"

[providers.kiro]
type = "kiro"
enabled = true
region = "us-east-1"

[logging]
level = "info"

CLI 命令

模型路由

模型按以下优先级路由：

1. 模型前缀 — copilot/gpt-4o 路由到 copilot Provider，使用模型 gpt-4o
2. X-Provider 请求头 — X-Provider: kiro 路由到 kiro Provider
3. 配置默认值 — 回退到配置文件中的 default_provider

API 端点

项目结构

src/
  auth/           # Token 管理
    copilot.rs    # GitHub 设备码 OAuth 流程
    kiro.rs       # Builder ID + 社交认证 (PKCE)
    token_store.rs
  cli/            # CLI 命令
  config/         # TOML 配置加载
  convert/        # OpenAI <-> Anthropic 格式转换
    stream.rs     # SSE chunk 转换
  daemon/         # 进程管理 + Unix Socket 控制服务
  providers/      # Provider 实现
    copilot/      # GitHub Copilot（VSCode 请求头伪装）
    kiro/         # Kiro / CodeWhisperer API
    router.rs     # 基于模型名的 Provider 路由
  routes/         # HTTP 路由处理器
    usage.rs      # 使用量跟踪端点
  server/         # Axum 服务器、中间件（认证、限流、CORS）
  usage/          # Token 使用量跟踪
    tracker.rs    # UsageTracker 实现
  util/           # XDG 路径、错误类型、流式辅助工具

文件位置

License

MIT