让每个模型都成为 Codex 引擎。

OpenAI 兼容的 Responses API 网关，让 Codex、CLI 工具和开发者 Agent 接入任意模型。

English Documentation · 中文文档

git clone https://github.com/Ahoo-Wang/GodeX.git
cd GodeX
bun install

npm install -g @ahoo-wang/godex
godex --help

docker pull ahoowang/godex:latest
# 或
docker pull ghcr.io/ahoo-wang/godex:latest

docker run -d \
  --name godex \
  -p 5678:5678 \
  -e ZHIPU_API_KEY=your-key \
  -e DEEPSEEK_API_KEY=your-key \
  -e MINIMAX_API_KEY=your-key \
  -e MIMO_API_KEY=your-key \
  -v ./godex.yaml:/etc/godex/godex.yaml:ro \
  -v godex-data:/data \
  ahoowang/godex:latest

godex init
godex serve --config ./godex.yaml

server:
  port: 5678
  host: 0.0.0.0

default_provider: deepseek

models:
  aliases:
    # -------------------------------------------------------------------------
    # Codex-compatible model aliases
    #
    # 这些 alias 是 GodeX routing policy，不代表与 OpenAI 原模型能力等价。
    # 依据优先级：公开 benchmark > 官方模型定位 > Provider 产品说明。
    # -------------------------------------------------------------------------

    # Codex 默认主力：复杂编码 / computer use / research workflows
    # 依据：DeepSeek V4-Pro 在 SWE / Terminal / Codeforces / GDPval-AA 上公开成绩强。
    gpt-5.5: "deepseek/deepseek-v4-pro"

    # Codex 旗舰：coding + reasoning + tool use + agentic workflows
    # 依据：DeepSeek V4-Pro 有更完整的公开 coding/agentic benchmark 覆盖。
    gpt-5.4: "deepseek/deepseek-v4-pro"

    # Codex mini：subagents
    gpt-5.4-mini: "zhipu/glm-5.1"

    # Codex 编码专用：复杂软件工程
    # 依据：DeepSeek V4-Pro 的 SWE Verified / SWE Pro / Terminal Bench 表现。
    gpt-5.3-codex: "deepseek/deepseek-v4-pro"

    # Codex spark：近实时编码迭代
    gpt-5.3-codex-spark: "zhipu/glm-5.1"

    # 上一代通用 coding / agentic fallback
    # 严谨起见仍走 DeepSeek；不强行映射到 Zhipu。
    gpt-5.2: "deepseek/deepseek-v4-pro"

    # -------------------------------------------------------------------------
    # Provider native models
    # -------------------------------------------------------------------------

    deepseek-v4-pro: "deepseek/deepseek-v4-pro"
    deepseek-v4-flash: "deepseek/deepseek-v4-flash"

    mimo-v2.5-pro: "xiaomi/mimo-v2.5-pro"
    mimo-v2.5: "xiaomi/mimo-v2.5"

    glm-5.1: "zhipu/glm-5.1"
    glm-5-turbo: "zhipu/glm-5-turbo"
    glm-4.7: "zhipu/glm-4.7"
    glm-4.5-air: "zhipu/glm-4.5-air"

    MiniMax-M3: "minimax/MiniMax-M3"

    # Fallback for unknown bare model names
    "*": "deepseek/deepseek-v4-pro"

providers:
  deepseek:
    spec: deepseek
    credentials:
      api_key: ${DEEPSEEK_API_KEY}
    endpoint:
      base_url: https://api.deepseek.com
  zhipu:
    spec: zhipu
    credentials:
      api_key: ${ZHIPU_API_KEY}
    endpoint:
      base_url: https://open.bigmodel.cn/api/coding/paas/v4
  minimax:
    spec: minimax
    credentials:
      api_key: ${MINIMAX_API_KEY}
    endpoint:
      base_url: https://api.minimaxi.com/v1
  xiaomi:
    spec: xiaomi
    credentials:
      api_key: ${MIMO_API_KEY}
    endpoint:
      base_url: https://api.xiaomimimo.com/v1

session:
  backend: sqlite

logging:
  level: info

trace:
  enabled: true
  path: ./data/trace.db
  capture_payload: false

godex serve --config ./godex.yaml

bun run dev

curl http://localhost:5678/health

curl http://localhost:5678/v1/models

curl http://localhost:5678/v1/responses \
  -H 'content-type: application/json' \
  -d '{
    "model": "gpt-5.5",
    "input": "写一个 TypeScript add 函数。"
  }'

curl -N http://localhost:5678/v1/responses \
  -H 'content-type: application/json' \
  -d '{
    "model": "gpt-5.5",
    "stream": true,
    "input": "用两句话解释 Bun streams。"
  }'

model = "gpt-5.5"
model_provider = "godex"

[model_providers.godex]
name = "GodeX"
base_url = "http://127.0.0.1:5678/v1"
wire_api = "responses"
requires_openai_auth = false
supports_websockets = false

bun install                  # 安装依赖
bun run dev                  # 热重载开发服务器，端口 13145
bun run start                # 从源码启动服务
bun run build                # 为当前平台编译二进制
bun run compile:all          # 交叉编译所有支持平台

bun run typecheck            # TypeScript
bun run lint                 # Biome check
bun run lint:fix             # Biome 自动修复
bun run format               # Biome 格式化
bun run test                 # 单元和集成测试，不含 src/e2e
bun run test:e2e             # mock 上游端到端测试
bun run test:zhipu           # 智谱 live 测试，需要 ZHIPU_API_KEY
bun run test:deepseek        # DeepSeek live 测试，需要 DEEPSEEK_API_KEY
bun run test:minimax         # MiniMax live 测试，需要 MINIMAX_API_KEY
bun run test:xiaomi         # Xiaomi live 测试，需要 MIMO_API_KEY
bun run check                # typecheck + lint + test
bun run ci                   # typecheck + biome ci + test + e2e

src/
  cli/          Commander CLI, init wizard, runtime config loading
  config/       godex.yaml schema, defaults, env interpolation
  context/      ApplicationContext and per-request ResponsesContext
  bridge/       Provider-agnostic Responses-to-Chat planning and reconstruction
  providers/    Built-in provider specs, hooks, clients, and registry
  responses/    Sync and stream request pipelines
  server/       Bun routes for /health, /v1/models, /v1/responses
  session/      Memory and SQLite response session stores
  trace/        SQLite trace recorder and usage/error/event mappers
  protocol/     OpenAI protocol type definitions
  error/        GodeXError hierarchy and domain codes

src/providers/<name>/
  spec.ts       ProviderSpec declaration
  client.ts     ProviderEdge construction with ChatProviderClient
  hooks.ts      Provider-specific patching, accessors, usage, stream deltas
  protocol/     Provider DTOs when needed
  index.ts      Public exports