v0.0.1

oc3

Name: oc3
Author: Griko Nibras

Run OpenCode Console models in OpenAI Codex and ChatGPT desktop — one local proxy, any Console model.

Install

npm install -g @nbrst/oc3

Or npx @nbrst/oc3 to try it, or curl -fsSL https://oc3.nbr.st/install | bash for a standalone binary (macOS arm64/x64, Linux x64/arm64).

Use

oc3 login   # device-code sign in to OpenCode Console
oc3 start   # proxy + config override + boots ChatGPT desktop
oc3 stop    # restores your previous config, stops the proxy

How it routes

responses — GPT/Grok/Muse models: near-passthrough, SSE relayed verbatim
chat-completions — full Responses translation: tool calls, reasoning, strict usage
messages / google — Anthropic and Gemini bridges
web_search — hosted tool bridged client-side via Exa/Parallel MCP
not in catalog — native OpenAI / ChatGPT fallback with the client's own auth

Also included

Schema-aware tool-argument repair, apply_patch envelope repair, strict usage details
Thinking/effort normalization per model family via a routing catalog
Auto-review alias resolution and Full-Access exec schema scrubbing for desktop
WebSocket attempts answered with 426 HTTP fallback; zstd request bodies accepted
Detached daemon with oc3 status / oc3 logs / one-command restore

How it routes

Also included

Links