[Feature] Official support for using Command Code plans as a provider in VS Code Copilot Chat’s model picker #1005

Feature Description

✨ The feature
Official support — via a first-party VS Code extension or a documented, stable provider API — for using Command Code plans (Go / GOAT / Pro / Max 10× / Max 20×) as selectable providers in VS Code Copilot Chat’s model picker, the same way this extension does for OpenCode: http://github.com/ltmoerdani/opencode-copilot-chat

🎯 The problem
Today, Command Code subscriptions are only usable inside the Command Code CLI and desktop app. That’s a real limitation for the growing group of developers whose daily workflow lives in VS Code Copilot Chat — Agent Mode, tool-calling, inline edits — and who want to keep that UX but swap the underlying model to their Command Code plan.

The community has already proven the pattern works on the OpenCode side: ltmoerdani/opencode-copilot-chat injects OpenCode Go/Zen models into Copilot Chat’s model picker via the Copilot Chat provider API, and it has a visible user base. Nothing equivalent exists for Command Code, and the gap shows up concretely:

Pro / Max subscribers can technically bridge via /provider/v1/*, but only through unofficial third-party proxies with no guarantees, no streaming-fidelity promises, and breakage on every CLI bump.
Go ($1/mo) subscribers are locked out entirely — /provider/v1/chat/completions and /provider/v1/messages return 403 upgrade_required, leaving only the undocumented /alpha/generate endpoint with a custom envelope that no official consumer outside the CLI uses.
There is no stable, documented contract for anyone building a bridge, so every third-party implementation re-derives the protocol on each release.
The demand is visible in the ecosystem: routers like decolua/9router and can1357/oh-my-pi are independently adding Command Code providers, reference protocol implementations (safzanpirani/pi-commandcode-provider) exist and get re-synced to each CLI version, and proxy projects keep popping up on GitHub
github.com
. All of that energy is currently fragmented because there’s no official, supported path.

💡 Proposed solution
Any of the following would work, in order of preference:

First-party VS Code extension. A small extension that registers Command Code models into Copilot Chat’s model picker using the same vscode.lm provider API that ltmoerdani/opencode-copilot-chat uses for OpenCode. Users sign in once (the existing user_... key from commandcode.ai/settings/billing would suffice — no OAuth needed, unlike Codex/Claude).
Officially document and stabilize the Provider API (/provider/v1/chat/completions, /provider/v1/messages, /provider/v1/models) and — ideally — extend it to the Go plan, or expose a thin OpenAI-compatible shim over /alpha/generate for Go users. With a documented contract, community extensions like the one above can add Command Code support themselves and stay in sync.
Minimum viable step: publish a reference implementation / protocol spec for /alpha/generate under a clear versioning policy, so bridges stop breaking on every CLI release.
Suggested UX if a first-party extension is pursued:

Command: Command Code: Sign In / Set API Key (stored in OS keychain)
Command: Command Code: Select Plan → surfaces only models the plan actually has access to, with per-model credit cost where available
Plan gating made explicit: Go users see a friendly “API access requires GOAT+” note rather than a raw 403
🔄 Alternatives considered
Third-party OpenAI-compatible proxies (HaeMeto/Command-Code-AI-Proxy, dev2k6/command-code-proxy-server, cmdcode2api): functional, but self-hosting a reverse-engineered gateway is fragile, doesn’t reliably handle streaming/usage details, and can’t serve Go-plan users at all since the custom envelope differs from OpenAI shape.
Staying in the CLI / desktop app: works, but I lose the Copilot Chat agent mode, tool-calling, and UI I already have wired into my VS Code workflow.
Switching to OpenCode Go: the ltmoerdani plugin would cover me, but Command Code’s model roster and per-model credit structure are why I subscribed — I’d rather stay and get official integration.
📋 Additional context
Community plugin proving the integration pattern works: http://github.com/ltmoerdani/opencode-copilot-chat
Command Code Provider API (OpenAI + Anthropic compatible, Pro+ only): https://commandcode.ai/docs/provider
Plans & pricing (Go / GOAT / Pro / Max 10× / Max 20×): https://commandcode.ai/pricing
Ecosystem demand evidence: router-for-me/CLIProxyAPI discussion #4007, can1357/oh-my-pi PR #1360, decolua/9router recent stream fixes
Related: the same feature request pattern exists for anomalyco/opencode #26338
Happy to test a beta extension or help validate the model-picker registration against a GOAT/Pro account.

Use Case

No response

Additional Context

No response

How important is this to you?

None