modelgo-cli

The official CLI for modelgo. Pairs with AI agent skills (Claude Code, Codex, Gemini CLI, etc.) so AI agents can operate modelgo on your behalf.

v0 framework stage. Device-flow auth commands are available. modelgo pay request supports anonymous x402 pay-per-call calls to model gateway APIs; API key management is not wired up yet.

Install

npx @model-go/cli@latest install

This runs an interactive wizard that:

1. Installs @model-go/cli globally via npm (which downloads the Go binary from GitHub Releases as modelgo).
2. Distributes modelgo-* skills to every AI agent installed on your machine (Claude Code, Trae, Trae CN, and other agents in the skills ecosystem).

After install, restart your AI agent (open a new chat / session) and try:

"Check my modelgo login status."

Your AI should find the modelgo-shared skill and run modelgo auth status.

Direct commands

modelgo --version
modelgo auth login
modelgo auth status
modelgo auth logout
modelgo pay methods
modelgo pay request --env cn --path /v1/chat/completions --method POST --data '{"model":"gpt-4o","messages":[]}' --json
modelgo --help

modelgo auth login without --no-wait is a blocking command: it prints the authorization URL, then keeps polling until the browser approval completes or the device code expires.

Environments

The CLI talks to a named environment. Two are built in:

| Env | Base URL | | ------ | ---------------------------- | | cn | https://api.modelgo.com | | intl | https://api.modelgo.ai |

The active environment defaults to cn. Switch it, or register your own:

modelgo env list                 # show built-in + custom envs (active marked with *)
modelgo env current              # print the active env
modelgo env use intl             # switch the active env
modelgo env add <name> --base-url https://...   # register or override an env URL
modelgo env remove <name>        # remove a custom env or override

Environment definitions and the active selection live in ~/.modelgo/config.json. Credentials are stored per-env in ~/.modelgo/auth.json, so switching environments preserves each login. There are no environment variables to set.

Auth commands operate on the active env; pass --env <name> to target a different one:

modelgo auth login --env intl
modelgo auth status --env intl
modelgo auth logout --all          # clear every env

Non-streaming agent flows

For AI-agent harnesses that cannot stream intermediate output to the user, use:

modelgo auth login --no-wait --json
modelgo auth login --device-code <DEVICE_CODE>

Do not show the URL and then immediately block on --device-code in the same turn. In final-message-only harnesses, that pattern prevents the user from ever seeing the URL before the agent starts waiting.

--json emits newline-delimited JSON (NDJSON). In the default waiting flow (modelgo auth login --json without --no-wait), stdout receives two JSON objects: first the device-code object (with verification_url), then the authenticated object once approved. Parse line by line — don't JSON.parse the whole stream. With --no-wait, only the device-code object is printed.

x402 pay-per-call model APIs

Use modelgo pay request when the user wants to call a model API without logging in to ModelGo first. The CLI opts into the gateway x402 path with X-Payment-Protocol: x402.

modelgo pay request --env cn --path /v1/chat/completions --method POST --data '{"model":"gpt-4o","messages":[{"role":"user","content":"hello"}]}' --intent "调用 gpt-4o 完成聊天补全" --json

In the cn environment, if the gateway returns a 402 that advertises alipay:*, the CLI writes the PAYMENT-REQUIRED payload to a local file and returns a structured handoff for alipay-payment-skill. In intl, the CLI does not route to Alipay; it returns the advertised x402 options for a future non-Alipay rail.

Upgrade

Re-run the installer; it detects an out-of-date install and upgrades in place:

npx @model-go/cli@latest install

License

MIT — see LICENSE.