{"id":22584,"date":"2010-05-14T08:15:07","date_gmt":"2010-05-14T08:15:07","guid":{"rendered":"https:\/\/scannn.com\/lidge-jun-opencodex-universal-provider-proxy-for-openai-codex-claude-code-use-any-llm-claude-gemini-grok-deepseek-ollama-with-codex-cli-app-sdk-and-claude-code-%c2%b7-gi\/"},"modified":"2010-05-14T08:15:07","modified_gmt":"2010-05-14T08:15:07","slug":"lidge-jun-opencodex-universal-provider-proxy-for-openai-codex-claude-code-use-any-llm-claude-gemini-grok-deepseek-ollama-with-codex-cli-app-sdk-and-claude-code-%c2%b7-gi","status":"publish","type":"post","link":"https:\/\/scannn.com\/lv\/lidge-jun-opencodex-universal-provider-proxy-for-openai-codex-claude-code-use-any-llm-claude-gemini-grok-deepseek-ollama-with-codex-cli-app-sdk-and-claude-code-%c2%b7-gi\/","title":{"rendered":"lidge-jun\/opencodex: Universal provider proxy for OpenAI Codex & Claude Code \u2014 use any LLM (Claude, Gemini, Grok, DeepSeek, Ollama\u2026) with Codex CLI, App, SDK, and Claude Code \u00b7 GitHub"},"content":{"rendered":"\n
\n

Universal provider proxy for OpenAI Codex & Claude Code<\/b> \u2014 use any LLM with Codex CLI, App, SDK, and Claude Code.<\/p>\n

npm install -g @bitkyc08\/opencodex<\/code> \u00b7 ocx start<\/code> \u00b7 localhost:10100<\/b><\/p>\n

\n <\/a>
\n \"license\"<\/a>
\n \"node<\/a>\n<\/p>\n

\n \"opencodex<\/a>\n<\/p>\n

\n English<\/a> \u00b7 \ud55c\uad6d\uc5b4<\/a> \u00b7 \u7b80\u4f53\u4e2d\u6587<\/a> \u00b7 \u0420\u0443\u0441\u0441\u043a\u0438\u0439<\/a> \u00b7 \u65e5\u672c\u8a9e<\/a> \u00b7 Full documentation \u2192<\/b><\/a>\n<\/p>\n

\n \"opencodex<\/a>\n<\/p>\n

Use Claude, Gemini, Grok, GLM, DeepSeek, Kimi, Qwen, Ollama, or any other LLM with Codex \u2014 and with Claude Code<\/strong> \u2014 without waiting for anyone to add support.<\/p>\n

opencodex is a lightweight local proxy that translates Codex’s Responses API into whatever your provider speaks. Streaming, tool calls, reasoning tokens, images \u2014 everything works, in both directions.<\/p>\n

\n \"opencodex<\/a>\n<\/p>\n

Codex, running any model.<\/b> Pick a provider and go \u2014 same Codex workflow, different brain.<\/sub><\/p>\n

It can also manage a ChatGPT account pool<\/strong> for Codex auth. Add multiple ChatGPT \/ Codex accounts,
\nrefresh their 5h \/ weekly \/ 30d quota in the dashboard, and let new sessions auto-route to the
\nlowest-usage healthy account. Existing Codex threads stay pinned to the account that started them,
\nso long SSH, tmux, or mobile-connected sessions do not jump accounts mid-conversation.<\/p>\n

\n
Codex CLI \/ App \/ SDK \u2500\u2500\/v1\/responses\u2500\u2500▶ opencodex \u2500\u2500▶ Any provider\n                                              \u2502\n              Anthropic \u00b7 Google \u00b7 xAI \u00b7 Kimi \u00b7 Ollama Cloud \u00b7 Groq\n              OpenRouter \u00b7 Azure \u00b7 DeepSeek \u00b7 GLM \u00b7 \u2026and OpenAI itself\n<\/code><\/pre>\n<\/div>\n
\n
\n
\n
flowchart LR\n  codex[Codex session<br\/>CLI, App, SSH, mobile] --> proxy[opencodex]\n  proxy --> existing{Existing thread?}\n  existing -->|yes| pinned[Keep the same<br\/>ChatGPT account]\n  existing -->|new session| quota[Refresh quota<br\/>5h, weekly, 30d]\n  quota --> pick[Pick lowest-usage<br\/>healthy account]\n  pick --> upstream[ChatGPT \/ Codex backend]\n  pinned --> upstream\n  upstream --> outcomes[Quota \/ auth outcome]\n  outcomes -->|429| cooldown[Cooldown + failover]\n  outcomes -->|401 \/ 403| reauth[Mark reauth needed]\n  cooldown --> quota\n<\/pre>\n<\/p><\/div>\n<\/p><\/div>\n
  
\n    
\n  Loading<\/span>
\n<\/span>
\n  <\/span>
\n<\/section>\n
<\/p>\n\n\n\n\n\n\n\n
OS<\/th>\nStatus<\/th>\nService manager<\/th>\n<\/tr>\n<\/thead>\n
macOS (arm64 \/ x64)<\/td>\nFully supported<\/td>\nlaunchd<\/td>\n<\/tr>\n
Linux (x64 \/ arm64)<\/td>\nFully supported<\/td>\nsystemd (user unit)<\/td>\n<\/tr>\n
Windows (x64)<\/td>\nFully supported<\/td>\nTask Scheduler (hidden) \/ opt-in native service (--native<\/code>, WinSW)<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
<\/markdown-accessiblity-table><\/p>\n
Requires Node<\/a> 18+. The Bun runtime is bundled automatically on npm install<\/code> \u2014 no separate Bun install needed. All three platforms work natively (no WSL needed on Windows).<\/p>\n
\n
#<\/span> Install (bundles the Bun runtime automatically \u2014 only Node 18+ required)<\/span>\n#<\/span> Prefer a user-owned Node (nvm\/fnm) \u2014 avoid `sudo npm install -g \u2026`<\/span>\nnpm install -g @bitkyc08\/opencodex\n\n#<\/span> Interactive setup (writes config, injects into Codex, and offers autostart shim install)<\/span>\nocx init\n\n#<\/span> Start the proxy<\/span>\nocx start\n\n#<\/span> If you skipped it during init, install the on-demand autostart shim later<\/span>\nocx codex-shim install\n\n#<\/span> Use Codex normally \u2014 it now routes through opencodex<\/span>\ncodex \"<\/span>Write a hello world in Rust\"<\/span><\/span><\/pre>\n<\/div>\n
\n“bundled Bun runtime is missing” \/ npm blocked Bun install scripts?<\/b><\/summary>\n
opencodex bundles the Bun runtime as a dependency and runs it via a Node
\nlauncher, so you do not<\/strong> need to install Bun yourself. If you see a
\n“bundled Bun runtime is missing” error, the install skipped lifecycle scripts
\n(including npm blocking bun’s postinstall under allowScripts<\/code>) or optional
\ndependencies. Reinstall without those flags, allowing bun’s install script:<\/p>\n
\n
npm install -g --allow-scripts=bun @bitkyc08\/opencodex   #<\/span> no --ignore-scripts, no --omit=optional<\/span>\n\n#<\/span> if the original install used sudo, keep using sudo:<\/span>\nsudo npm install -g --allow-scripts=bun @bitkyc08\/opencodex<\/pre>\n<\/div>\n
npm’s own warning suggests an abbreviated command without the package name \u2014
\nthat would reinstall the current directory, so always pass
\n@bitkyc08\/opencodex<\/code> explicitly.<\/p>\n
If you installed with sudo<\/code> into a root-owned prefix, the sudo reinstall above
\nunblocks that prefix \u2014 but prefer migrating to a user-owned Node (nvm, fnm, or
\na user npm prefix) when you can.<\/p>\n<\/details>\n
The fastest way to add a provider is through the web dashboard:<\/p>\n
This opens the dashboard at http:\/\/localhost:10100<\/code>. From there:<\/p>\n
    \n
  1. Click “Add Provider”<\/strong><\/li>\n
  2. Pick from 40+ built-in providers<\/strong> \u2014 or enter a custom OpenAI-compatible endpoint<\/li>\n
  3. Paste your API key (or log in via OAuth for Anthropic, xAI, and Kimi)<\/li>\n
  4. Models are auto-discovered<\/strong> from the provider’s \/v1\/models<\/code> endpoint<\/li>\n<\/ol>\n
    Your new provider is ready to use immediately. No restart needed.<\/p>\n
    You can also add providers through ocx init<\/code> (interactive CLI) or by editing ~\/.opencodex\/config.json<\/code> directly.<\/p>\n
    Target any configured provider and model using the provider\/model<\/code> syntax:<\/p>\n
    Providers whose own model ids contain \/<\/code> (zenmux, openrouter, nvidia, \u2026) are exposed to
    \nCodex with inner slashes aliased to -<\/code> (e.g. zenmux\/moonshotai-kimi-k3-free<\/code>); the
    \nproxy transparently routes them back to the native id, and the raw full-slash form keeps
    \nworking too.<\/p>\n
    \n
    #<\/span> Use Claude Opus through Anthropic<\/span>\ncodex -m \"<\/span>anthropic\/claude-opus-4-8\"<\/span><\/span> \"<\/span>Explain this stack trace\"<\/span><\/span>\n\n#<\/span> Use Gemini through Google<\/span>\ncodex -m \"<\/span>google\/gemini-3-pro\"<\/span><\/span> \"<\/span>Write unit tests for auth.ts\"<\/span><\/span>\n\n#<\/span> Use GLM through Ollama Cloud<\/span>\ncodex -m \"<\/span>ollama-cloud\/glm-5.2\"<\/span><\/span> \"<\/span>Write a SQL migration\"<\/span><\/span>\n\n#<\/span> Use a local model through Ollama<\/span>\ncodex -m \"<\/span>ollama\/llama3\"<\/span><\/span> \"<\/span>Refactor this function\"<\/span><\/span><\/pre>\n<\/div>\n
    When you omit the provider\/<\/code> prefix, opencodex routes to the default provider \u2014 or auto-matches based on the model name pattern (e.g., claude-*<\/code> routes to Anthropic, gpt-*<\/code> routes to OpenAI).<\/p>\n
    Routed models also appear in the Codex App<\/strong> model picker with per-model reasoning effort controls:<\/p>\n
    Current Codex builds can expose low<\/code>, medium<\/code>, high<\/code>, xhigh<\/code>, max<\/code>, and ultra<\/code> reasoning
    \ncontrols when a model advertises them. opencodex keeps xhigh<\/code> and max<\/code> distinct unless a provider
    \nconfig explicitly maps one to the other. ultra<\/code> mirrors upstream Codex semantics: it selects
    \nmaximum reasoning plus proactive multi-agent delegation in the client, and is converted to max<\/code>
    \nbefore any request reaches a provider. Routed models advertise it only when a provider config opts
    \nin via reasoningEfforts<\/code>.<\/p>\n
    GPT-5.6 Sol\/Terra\/Luna are seeded as rollout-ready catalog entries for the OpenAI API key and
    \nOpenRouter presets (gpt-5.6-sol<\/code>, gpt-5.6-terra<\/code>, gpt-5.6-luna<\/code>; OpenRouter uses
    \nopenai\/...<\/code>). They remain preview-gated by upstream availability; opencodex only prepares the
    \nrouting and catalog metadata for accounts and providers that can serve them.<\/p>\n
    \n  \"Codex<\/a>\n<\/p>\n
    \n
    OpenAI provider account modes<\/h2>\n
    <\/svg><\/a><\/div>\n
    <\/p>\n\n\n\n\n\n\n
    Provider ID<\/th>\nRoute<\/th>\nCredential<\/th>\nBehavior<\/th>\n<\/tr>\n<\/thead>\n
    openai<\/code><\/td>\nCodex login<\/td>\nMain + added Codex accounts<\/td>\nPool by default; optional Direct mode<\/td>\n<\/tr>\n
    openai-apikey<\/code><\/td>\nOpenAI API<\/td>\nAPI key\/key pool<\/td>\nNo Codex account routing<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
    <\/markdown-accessiblity-table><\/p>\n