subscriptions

use a plan you already pay for, ChatGPT Plus or Pro, or a coding plan from Z.ai, Kimi or MiniMax, in place of an API key.

an API key bills each call. a plan is a monthly price you may already pay. bise can use a few plans, signed in with the provider's own login or with the plan's key.

the first run asks how you want to pay for the models: Continue with ChatGPT, OpenRouter, an API key, or a coding plan key. later, /provider adds or changes any of them.

#ChatGPT Plus or Pro

bise signs in with OpenAI's "Sign in with ChatGPT": your ChatGPT plan pays for the calls, not an API account.

#sign in

in bise, the first run's Continue with ChatGPT, or /provider, then ChatGPT. in a terminal:

bise login chatgpt

opening your browser to sign in to ChatGPT…
or open this link:
  https://auth.openai.com/...
waiting… ctrl+c cancels.
✓ signed in as you@example.com · ChatGPT Plus.

sign in to ChatGPT in the browser and allow bise to use your plan. bise checks the plan with one tiny call, then main and your agents use a ChatGPT model on your plan.

#over SSH, or from an agent

bise login chatgpt --no-browser

it prints the link and waits. the sign-in comes back to 127.0.0.1 on the machine where bise runs, so over SSH, forward that port first; the command prints the exact line, like ssh -L <port>:127.0.0.1:<port> <host>.

a coding agent setting bise up for you can run this command, but only you can sign in: the browser asks for your consent.

#the models

ChatGPT models are named chatgpt/<model>, for example chatgpt/gpt-6.1-sol. they are OpenAI's models, billed to your plan. openai/<model> stays your API key: you can have both and mix them across roles.

/models lists the models your plan offers, and says your ChatGPT plan where the others show a price.

bise config set main chatgpt/gpt-6.1-sol
bise config set small chatgpt/gpt-6-luna

#limits and errors

your plan has a usage limit for apps like bise. when it's reached, the turn ends with:

▲ your ChatGPT plan's limit for bise is reached. your usage is at chatgpt.com/settings/usage, or switch model with /model.

your usage is at chatgpt.com/settings/usage; /provider links it too. the other lines you may see:

#sign out

bise logout chatgpt

bise tells ChatGPT to end its sign-in. if ChatGPT doesn't confirm, the line says so: then remove bise in your ChatGPT settings to be sure.

#OpenRouter

bise login openrouter            # asks: sign in with your browser, or paste a key
bise login openrouter --browser
bise login openrouter --key
bise login openrouter --no-browser   # prints the link (SSH: forward its port, as for ChatGPT)

the browser sign-in makes a normal OpenRouter API key for bise and saves it, so nothing else changes: the models are openrouter/<vendor>/<model>, paid with your OpenRouter credit.

#coding plans

Z.ai, Kimi and MiniMax sell coding plans for tools like bise. each plan gives you an API key for its coding endpoint; bise has a provider for each.

bise login zai-coding

or the first run's a coding plan key. the key is checked with one tiny call, then saved like any key (see providers and keys).

#where it's saved

the ChatGPT token renews by itself while bise runs. a token never goes in the environment, a log, or a command line.

bise auth status        # every provider: how it logs in, its state. never a secret
bise auth status --json # the same, for a script or an agent
bise doctor             # also says how long the ChatGPT sign-in is good for

#what bise sees from your other tools

bise looks whether Codex is signed in with ChatGPT, and whether Claude Code is signed in with a Claude plan. it looks only at whether their login is there (a file, or on a Mac a keychain entry), never at the tokens, to tell you the right next step:

• Codex with ChatGPT: the first run marks Continue with ChatGPT, "you use it in Codex already". bise still signs in on its own.
• Claude Code with a Claude plan: one line says that plan works only in Claude Code.

#not supported