commandcode-go-opencode

Unofficial bridge that lets you use your Command Code Go plan from OpenAI- and Anthropic-compatible clients — OpenCode, Claude Code, Continue.dev, curl, or anything that speaks POST /v1/chat/completions.

⚠️ Disclaimer

This project is not affiliated with, endorsed by, or supported by Command Code. By using it, you accept that:

• Your Command Code account may be suspended or terminated without notice. The maintainers of this project cannot help you recover a suspended account — only support@commandcode.ai can.
• The /alpha/generate endpoint used here is the same one the official Command Code CLI uses. It is not part of any documented public API, and Command Code may change or rate-limit it at any time.
• This tool exists because the official Provider API is gated behind the paid Provider plan, while the 1ドル/mo Go plan is CLI-only. If Command Code decides that's not OK, the right place to take that up is with them, not by harassing this project.

If you don't accept those terms, don't install this tool. Use the official CLI (npm i -g command-code) or upgrade to the Provider plan.

The Command Code Go plan (1ドル/month) is the cheapest way to run a coding agent in a terminal, but it has one sharp edge: the model catalogue is only reachable from the official CLI. The OpenAI-compatible Provider API (https://api.commandcode.ai/provider/v1/chat/completions) is gated behind the paid Provider plan, and the upstream returns {"code": "upgrade_required"} if you try it on Go.

This tool bridges that gap. It runs a small local HTTP server on 127.0.0.1:8787, accepts OpenAI / Anthropic requests from your favourite client, and forwards them to Command Code's /alpha/generate endpoint — the same one the official CLI uses. From Command Code's perspective, your traffic looks like normal CLI usage; from your client's perspective, it sees a normal OpenAI-compatible server.

npm install -g commandcode-go-opencode

Or, to run from source without installing globally:

git clone https://github.com/lordon1a/commandcode-go-opencode
cd commandcode-go-opencode
npm install
npm run build

Open https://commandcode.ai/settings/keys and create a key. The CLI and the API share the same key.

commandcode-go-opencode auth login
# paste the key when prompted

Or non-interactive:

commandcode-go-opencode auth login --key "user_..."

The key is stored at ~/.config/commandcode-go-opencode/auth.json with 0600 permissions. Any CR/LF/NUL/whitespace characters that snuck into the key during paste are stripped before saving — this protects against a known bug where PowerShell copy-paste leaves carriage returns in environment variable values, which would otherwise cause upstream to reject the request.

commandcode-go-opencode serve
# [bridge] listening on http://127.0.0.1:8787

It stays in the foreground; stop it with Ctrl+C. Bind only to 127.0.0.1 — never expose this to the network. There is no client-side authentication: it's designed to be reachable only from your own machine.

commandcode-go-opencode setup-opencode --write
# Wrote provider 'commandcode-go' to ~/.config/opencode/opencode.jsonc

This adds a commandcode-go provider entry to your OpenCode config, pointing at the bridge on http://127.0.0.1:8787/v1. Other providers you have configured are preserved.

Then in OpenCode, run /model and pick any model under the commandcode-go heading — deepseek-v4-flash is a good default.

For Claude Code, the bridge exposes an Anthropic-compatible POST /v1/messages endpoint. Set:

export ANTHROPIC_BASE_URL=http://127.0.0.1:8787
export ANTHROPIC_AUTH_TOKEN=proxy-managed # value is unused, just non-empty
export ANTHROPIC_DEFAULT_SONNET_MODEL=deepseek-v4-pro
export ANTHROPIC_DEFAULT_OPUS_MODEL=deepseek-v4-pro
export ANTHROPIC_DEFAULT_HAIKU_MODEL=deepseek-v4-flash

Or run the generator script:

# macOS / Linux
./scripts/setup-claude-code.sh
# Windows (PowerShell)
./scripts/setup-claude-code.ps1

curl -s http://127.0.0.1:8787/v1/models | head -c 200
# {"object":"list","data":[{"id":"deepseek-v4-flash","object":"model",...}]}
curl -s http://127.0.0.1:8787/v1/chat/completions \
 -H "Authorization: Bearer proxy-managed" \
 -H "Content-Type: application/json" \
 -d '{
 "model": "deepseek-v4-flash",
 "messages": [{"role": "user", "content": "Say OK only."}]
 }'
# {"choices":[{"message":{"role":"assistant","content":"OK"},...}]}

The bridge ships the Go plan's model list. Anything not on this list will be rejected locally with unsupported_model before it hits the upstream — saving you from accidentally spending credits on a Provider-plan-only model that would 403 anyway.

Aliases are case-insensitive and the upstream id also works. See src/models.ts for the canonical list. If Command Code rotates the list, send a PR updating that file.

Premium Provider-plan models — Claude (all), GPT-5.6 Sol/Terra, Gemini 3.5+, Muse Spark 1.1 and standard 1.2/1.3 — are not available here, even if you pass their ids.

commandcode-go-opencode <command> [options]
Commands:
 auth login [--key <key>] Store API key. Interactive if --key omitted.
 auth logout Remove stored key.
 auth status Show whether a key is stored (no value printed).
 serve [--port N] [--host H] Start the proxy. Default 127.0.0.1:8787.
 setup-opencode [--write] Print or write OpenCode provider config.
 models List Go plan models.
 help Print this help.
Get your API key from https://commandcode.ai/settings/keys.

The bridge has no other configuration. It does not auto-update, phone home, or write telemetry. The only outbound request it makes per chat is the forwarded request to api.commandcode.ai.

• Single concurrent client. Don't point OpenCode, Claude Code, and curl at the bridge at the same time. Concurrent sessions may trip the same anti-abuse heuristics the CLI does.
• Match your normal CLI cadence. If you normally send 30 requests a day, don't suddenly send 300. Tool-call-heavy workflows look like agent traffic from upstream's perspective.
• Don't run multiple machines. Two machines sharing one API key is a classic account-takeover pattern. Use one key per machine.
• Rotate the key if you suspect leakage. auth login --key ... overwrites the file. Old key is invalidated as soon as Command Code issues the new one.
• Don't disable ZDR. Command Code's CMD_ZDR=1 / x-cmd-zdr: 1 zero-data-retention header is honoured by upstream. If you care about privacy, keep it on.

 ┌────────────────────────────────┐
 │ OpenCode / Claude Code / curl │ OpenAI or Anthropic
 │ on your machine │ /v1/chat/completions
 └─────────────┬──────────────────┘ /v1/messages
 │
 ▼
 ┌─────────────────────────────────────────┐
 │ commandcode-go-opencode (this project) │
 │ 127.0.0.1:8787 │
 │ - Validates model id │
 │ - Translates OpenAI ↔ /alpha/generate │
 │ - Streams SSE both ways │
 └─────────────────┬───────────────────────┘
 │
 ▼
 https://api.commandcode.ai/alpha/generate
 (the same endpoint command-code CLI uses)

The bridge is intentionally tiny. It does not implement tool-call negotiation (clients that need it should use the official CLI). It does not implement the full Provider API surface — only the parts that clients actually use.

The bridge stripped CR/LF before saving, so this shouldn't happen with this tool. If you see it, your key is being injected by something other than the bridge (a global env var? a shell function?). Re-run auth login and remove any duplicate key sources.

You might be on the Go plan but the upstream sometimes returns this for edge cases. Verify with auth status that your key is stored, then curl -s -H "Authorization: Bearer proxy-managed" http://127.0.0.1:8787/v1/models. If v1/models works but a chat request doesn't, it's an upstream issue, not the bridge.

Run commandcode-go-opencode models to see the bundled catalogue. If it's empty, your npm install was incomplete — try reinstalling.

Please read the disclaimer. Only support@commandcode.ai can help. Be polite, explain that you were using a third-party CLI tool, and ask whether you can be reinstated. There is no guarantee. Do not open a public issue here demanding that we "fix it" — we cannot.

See CONTRIBUTING.md. Run npm test before sending a PR. The CI matrix tests Node 18, 20, and 22 on Ubuntu, macOS, and Windows.

MIT. See LICENSE.