opencodex

Why we included this project

The official Codex and Claude Code CLIs assume you'll stick with the model each vendor ships. opencodex removes that constraint with a small local proxy that translates the Responses API those tools expect into whatever your chosen provider speaks, handling streaming, tool calls, reasoning tokens, and images in both directions. A single install lets you point Codex at Gemini or Ollama, or run Claude Code on a non-Anthropic model, while keeping the stock interface and the sub-agent picker; Grok Build and Claude Desktop can be pointed at other providers the same way. Beyond simple routing, it adds practical operations features: a web dashboard for managing providers and models, failover and weighted round-robin combos that fold several providers into one virtual model id, and quota-aware pooling of ChatGPT accounts for Codex auth so long sessions don't jump accounts. Teams that standardize on one coding CLI but want flexibility over backends, cost, or local models will get the most out of it.