Which AI models can I use?

The same catalog as your GlobalGPT subscription — GPT-class chat models, Gemini, DeepSeek, Kimi and more, plus image models like Nano Banana Pro and video models like Seedance and Sora. The model list is served by the backend, so new models appear without updating the CLI.

How does billing work?

It's your normal GlobalGPT balance — the same credits as the web app, nothing separate to top up. Media generation always shows a cost estimate and asks for confirmation before spending, and web search bills a small flat fee per search.

Do I need an API key?

No. glbgpt login opens your browser to authorize; on a remote or headless machine, glbgpt login --code lets you paste a one-time code instead.

Does it work with Claude Code, Codex, or Cursor?

Yes, three ways: paste the remote MCP URL https://mcp.glbgpt.com/mcp into Claude Code, Claude.ai, ChatGPT or Codex and sign in with your account (no install); glbgpt mcp runs the same tools as a local stdio server; and glbgpt skill install adds Agent Skills so coding agents can generate media or delegate subtasks through your account.

Can an agent's spending run away?

No. Unattended media generation is capped by a per-session credit budget (4000 by default); set --max-credits to adjust it, or --max-credits 0 to disable media spend entirely — useful in CI.

Can it read my files? Is that safe?

Tool-capable models can list, read, and search files under your working directory — strictly read-only, and file contents are only sent to the model when it asks to read them. The CLI never edits your files.

🎉 Celebrate 2 Years with Dela AI. Built for work, creation, and discovery. Get 50% OFF.

Compare plans

FLASH SALE

Save 50% on

Yearly Plan

Grab The Deal

GLOBALGPT CLI

The all-in-one AI toolbox, now in your agent/terminal

Chat with 100+ AI models, generate images and video — your GlobalGPT account in the terminal. Sign in with your account, no API key; billing is your normal balance. Requires Node.js ≥ 18.

$npm install -g @glbgpt/cli

then

$glbgpt login

→$glbgpt

Claude Code

Codex

Cursor

OpenClaw

Any MCP client

Scripts & CI

glbgpt — ~/projects

GlobalGPT CLI

Chat, images, and video — your GlobalGPT account in the terminal.

model: gpt-5.6-sol

plan: Pro · balance: 12,480 credits

› Summarize error.log — what caused the 02:14 spike?

● Read error.log (14 MB). It's a retry storm: the payment

webhook returned 503 and the queue re-enqueued without

backoff. Two fixes below — want a patch?

/model to switch · @./photo.png to attach · Esc interrupts

GUIDES & DEEP DIVES

How to Use GlobalGPT CLI in Cursor for Cheap AI

Add chat, image, and video generation to Cursor through one subscription — setup in minutes.

Read the guide →

How to Use GlobalGPT CLI in Claude Code

Wire the CLI into Claude Code with MCP and Agent Skills — images, video, and delegation from your agent.

Read the guide →

How to Use GlobalGPT CLI in Codex

Connect the CLI to Codex over MCP — one subscription for chat, images, and video inside your agent.

Read the guide →

THE ENTIRE TOOLKIT

One CLI, every workflow

GETTING STARTED

From npm to a first answer in a minute

One command installs glbgpt. Sign in through the browser — no API key — and the TUI opens onto your working directory: tool-capable models can list, read, and grep your files, read-only, to answer questions about them.

glbgpt login --code

glbgpt resume last

@./photo.png

zsh — ~/dev/shop

$ npm install -g @glbgpt/cli

added 92 packages in 6s

$ glbgpt login

Opening your browser to authorize…

Logged in as dana@studio.com.

$ glbgpt

› what changed under src/ since v0.4? keep it short

● Read 12 files. Three changes matter: the retry queue,

the token refresh, and the new /tasks command.

Fresh install → browser sign-in → asking about local files.

glbgpt-output/Claude Code

White sneaker on wet asphalt at night, studio light — generated with the glbgpt CLI

glbgpt-nano-banana-pro.png · 2048×2048~500 credits

The file lands in glbgpt-output/ inside your repo.

SKILLS · IMAGE

Ask your agent for an image — the skill does the rest

After glbgpt skill install, "make me a product shot" triggers the globalgpt-image skill: it prices the job first, puts the reference in -i — never the prompt — and runs glbgpt image.

Make a 1:1 product shot: white sneaker on wet asphalt, studio light

Nano Banana Pro1:1 · 2K

⏺ Estimated ~500 credits — confirmed. Saved glbgpt-output/glbgpt-nano-banana-pro.png.

glbgpt skill install

--estimate first

SKILLS · VIDEO

A still becomes a clip while Codex stays on task

The globalgpt-video skill handles image-to-video: your start frame goes in --start-image, the render takes minutes and blocks until saved — and a failed render is never retried automatically, because interrupted renders may still be charged.

Turn the sneaker shot into a 5s slow orbit with soft depth of field

Seedance 2.016:9 · 5s

⏺ Estimated ~1200 credits — confirmed. Rendering takes minutes… saved glbgpt-output/glbgpt-seedance-2.0.mp4.

task wait last

--duration 5

glbgpt-output/Codex

glbgpt-seedance-2.0.mp4 · 5.0s~1200 credits

Rendered in the cloud, delivered next to your code.

glbgpt model list chat — pick a delegate

$ glbgpt exec "add JSDoc to every exported function" \ --file src/utils/date.ts -m deepseek-v4-pro

Pick a model — the exec command updates. The choice is the point.

SKILLS · CODING DELEGATE

Delegate a subtask to the model you pick

/globalgpt-coding lists your account's chat models as a menu. The delegate gets read-only file tools jailed to the repo and produces text only — your agent reviews and applies it. You stay the engineer.

/globalgpt-coding

⏺ 6 functions documented by the delegate — reviewing, then applying.

MCP · ANY CLIENT

No shell? Paste one URL — with a hard budget

Remote MCP at mcp.glbgpt.com: add the URL to Claude Code, Claude.ai, ChatGPT or Codex and authorize in the browser — no install, no API key. Unattended spend is capped per session — past the cap the tool refuses and nothing is charged. A local stdio server (glbgpt mcp) ships in the CLI for local-file workflows.

Generate a 16:9 hero image with GlobalGPT

⏺ glbgpt_generate_image → glbgpt-output/hero.png · ~500 credits

mcp.glbgpt.com/mcp

--transport http

OAuth sign-in

mcp.glbgpt.com — remote MCP● connected

glbgpt_chatone turn: question in, answer out

glbgpt_generate_imagefinished result in one call

glbgpt_generate_videoreturns a task_id to check

glbgpt_check_taskwaits server-side, call again until done

glbgpt_list_modelswhat each model supports

session budget3,996 / 4,000 credits left

--max-credits <n> to adjust · 0 disables media entirely

Five tools, one registration line, budget enforced.

release.yml — nightlyexit 0

✔Set up job0.4s

✔npm install -g @glbgpt/cli6.2s

✔glbgpt exec "summarize release notes" --json --max-credits 02.1s

{"type":"result","model":"gpt-5.6-sol",

"content":"Faster retries and a new /tasks view. 12 fixes.","error":null}

--max-credits 0 guarantees a CI run can never spend on media.

EXEC · SCRIPTS & CI

Pipe it like any Unix tool

glbgpt exec never opens the TUI: prompt in, answer on stdout, exit 0 on success. --json for one result object, --output-schema for structured output, --jsonl for streaming events.

--jsonl

--output-schema

--max-credits 0

Frequently asked questions

A terminal client for your GlobalGPT account: chat with 100+ AI models, generate images and video, and pipe results into scripts. One npm install, sign in with your account — no API key to manage.

Install once. Create anywhere.

No API key — sign in with your GlobalGPT account. Requires Node.js ≥ 18.

$ npm install -g @glbgpt/cli