claude-opencode-go

Run Claude Code locally using OpenCode Go models.

Claude Code is a premier agentic harness, but is native to Anthropic endpoints. claude-opencode-go bridges Claude Code CLI with OpenCode Go models (glm-5.2, minimax-m3, qwen3.7-max, kimi-k3, grok-4.5) etc.

This CLI spins up a zero-trust local proxy, sanitizes incoming payloads in real-time, injects your OpenCode key, and streams SSE responses back with zero TTFT latency buffering. Note: While Claude Code's interactive /model picker UI is locked to Anthropic defaults, model selection is specified directly via CLI flag (-m).

Navigation: Install · Usage · Terminal Integration · Proxy Testing · Bruno Collection · Security · Agents

install

From Source (Local Development)

git clone https://github.com/ashwingopalsamy/claude-opencode-go.git
cd claude-opencode-go
npm install
npm run build
npm link

From NPM

npm install -g claude-opencode-go

Requires Node 18+ and Claude Code (npm install -g @anthropic-ai/claude-code).

usage

opencode-claude

On first run, it auto-detects your key from ~/.local/share/opencode/auth.json or prompts for it. Keys stay local in ~/.opencode-claude/config.json.

Specify a model:

opencode-claude -m qwen3.7-max

terminal integration (~/.zshrc & ~/.claude/settings.json)

To launch OpenCode models seamlessly whenever you type claude in your terminal:

1. Shell Alias (~/.zshrc or ~/.bashrc)

Add an alias to your ~/.zshrc (zsh) or ~/.bashrc (bash):

alias claude="opencode-claude"

Or pin your favorite model directly:

alias claude="opencode-claude -m glm-5.2"

Reload your shell profile:

source ~/.zshrc

2. Environment Variables (~/.zshrc)

Optionally set a default model via environment variables:

export OPENCODE_MODEL="glm-5.2"

3. Claude Code Settings (~/.claude/settings.json)

Configure fallback defaults in ~/.claude/settings.json:

{
  "model": "glm-5.2",
  "availableModels": [
    "glm-5.2",
    "minimax-m3",
    "qwen3.7-max",
    "kimi-k3",
    "grok-4.5"
  ],
  "enforceAvailableModels": true
}

[!NOTE] Model Picker Limitation: The availableModels list in ~/.claude/settings.json currently does not populate models inside Claude Code's interactive /model picker UI. You must specify your desired model ID manually via the -m flag (e.g. opencode-claude -m qwen3.7-max) or set "model" in ~/.claude/settings.json.

standalone proxy testing

opencode-claude --proxy-only --port 4000

Point Bruno, Postman, or cURL to http://127.0.0.1:4000/v1/messages.

bruno collection

A ready-to-use Bruno API collection is available under docs/bruno for local proxy testing and upstream verification:

• GET Models (Discovery) — test /v1/models endpoint
• POST Messages (Proxy Test) — test /v1/messages payload stripping & SSE streaming with proprietary Claude Code headers
• POST Messages (Direct Upstream) — direct benchmark against https://opencode.ai/zen/go/v1/messages

Includes pre-configured environments for Local Proxy (http://127.0.0.1:4000) and Direct OpenCode.

license

MIT