@bacnh85/pi-commandcode

Pi extension that connects to Command Code's OpenAI-compatible Provider API.

Install

pi install npm:@bacnh85/pi-commandcode

What it does

Registers the commandcode provider in Pi. Models are fetched dynamically from GET /provider/v1/models and appear in Pi's /model picker under the commandcode/ prefix. Pi's built-in OpenAI completions client handles all streaming and tool calling — no custom API code.

Usage

Log in

/login commandcode

This uses Pi's built-in /login flow — no dedicated slash command needed. Paste your Command Code API key (generate one in Studio → API Keys).

Alternatively, set the API key via environment variable (for CI/headless):

export COMMAND_CODE_API_KEY=cmd_...

Once logged in, models populate via catalog refresh and /model shows them:

/model commandcode/deepseek/deepseek-v4-flash

Environment variables

| Variable | Default | Description | |----------|---------|-------------| | COMMAND_CODE_API_KEY | — | API key (overrides /login; read by catalog refresh) | | COMMAND_CODE_BASE_URL | https://api.commandcode.ai/provider/v1 | Override the base URL (must include /v1; chat completions POST to <baseUrl>/chat/completions) |

settings.json (base URL)

The endpoint URL lives in settings.json under commandcode.baseUrl (precedence: env > repo .pi/settings.json > global ~/.pi/agent/settings.json > default). Edit it interactively:

/commandcode-config

Arrow keys navigate, Enter edits, Esc saves — the panel re-registers the provider at the new endpoint and refreshes the model catalog; pick models with /model. In non-TUI mode (or with the show arg) the command prints a config summary. The API key is never stored in settings.json — it lives in auth.json via /login commandcode or the env var.

How it works

1. On startup, registers the commandcode provider with api: "openai-completions".
2. Models are loaded from a disk cache (instant session restore) and warmed in the background from /provider/v1/models.
3. After /login, Pi's catalog refresh calls refreshModels, which fetches the live model list authenticated with your stored key.
4. Pi's built-in OpenAI completions API handles all streaming and tool calling.

The Provider API bills at model cost with no markup. See Command Code's Provider API docs and Pricing & Limits for plans, credits, and per-token rates.

Notes

• API key vs. CLI subscription: This extension uses the Provider API surface (OpenAI-compatible, pay-as-you-go or plan credits). The Command Code CLI subscription (GOAT/Pro/Max) and its /usage rolling windows are tracked on the web Studio and are not exposed via the Provider API key.
• Usage display: Install @bacnh85/pi-sub for a session-cost footer when a commandcode model is active. Because the Provider API key does not expose live usage windows, pi-sub shows session cost and tok/s (same as opencode-go and 9router), not 5h/weekly meters.

Changelog

See CHANGELOG.md for release history.

License

MIT