Ocincc — AI skill for Claude Code

ocincc — OpenCode Zen free models inside Claude Code

Run [OpenCode Zen](https://opencode.ai/zen)'s **free AI models** (Big Pickle, Ox Alpha, MiMo, Hy3, Nemotron, DeepSeek Flash, Laguna…) inside **Claude Code** through a local protocol-translating proxy.

npx ocincc@latest        # from the npm registry
# or, directly from GitHub:
npx github:farseenmanekhan1232/ocincc

That's it. The installer sets up a local LiteLLM proxy and a `zen-claude` launcher. Your existing `claude` command and `~/.claude/settings.json` are never touched.

---

Why this exists

Claude Code only speaks the **Anthropic Messages API** (`/v1/messages`). Most of Zen's free models are only served via the **OpenAI chat/completions** format, so pointing Claude Code directly at `https://opencode.ai/zen` fails for them.

`ocincc` runs a tiny local proxy that:

Claude Code ──(Anthropic /v1/messages)──▶ localhost:4000 ──translate──▶ https://opencode.ai/zen/v1/chat/completions

What you get

• ✅ All current free Zen models in Claude Code (auto-discovered at launch)
• ✅ Interactive model picker every session (or --model )
• ✅ Context window auto-configured per model (CLAUDE_CODE_MAX_CONTEXT_TOKENS) — 1M-context models get their full window
• ✅ Automatic 429 fallbacks between models
• ✅ Fresh client headers per session so you get the standard free-tier allowance
• ✅ Zero changes to your Claude Code config

Requirements

• macOS or Linux
• Claude Code (claude on PATH)
• Python 3 + curl
• A free OpenCode API key from opencode.ai/auth — no credit card needed for free models

Install & use

npx ocincc@latest          # one-time setup (prompts for your API key)
zen-claude                 # pick a model → start working

Launcher commands:

Command	Description
zen-claude	Interactive model picker, then launches Claude Code
zen-claude --model big-pickle	Skip the picke