Models and keys

A profile’s welcome asks for a model the first time. As the admin you can add or replace API keys, sign in with ChatGPT and add models on the web, under Models & keys in your account menu, or from the terminal as below. Model presets are shared by all profiles.

The nolune plan

A subscription to nolune itself: chats, pictures and memory search, with no API keys to get. It’s $20 a month for $25 of credits while the launch offer lasts, spent at what OpenRouter charges for each request; see Pricing and the terms.

1. Subscribe on your account page, which signs you in with a code sent to your email.
2. Link nolune to it: pick nolune plan in a new profile’s welcome, Link on its row in Models & keys, or run nolune nolune-plan setup. Each shows a code and a link page; open it on any device, sign in, check that it shows the same code, and link it. A phone works as well as this computer.
3. Pick a model. The plan offers the models OpenRouter serves that can call tools, with Claude Sonnet 5.5 first; nolune nolune-plan models lists them, and nolune preset add anthropic/claude-sonnet-5.5 --provider nolune-plan adds one.

Pictures on the Images page and search by meaning then use the plan too, unless you’ve set a key of your own for them. The whole family draws on one plan.

• Credits. Each month’s start brings new ones, and up to $12.50 of what’s left carries over. Automations and other background work can spend a tenth of the month’s credits in a day, so one that runs away can’t spend the month by itself. How much is used shows in the model menu, under the message box as it runs low, and in Models & keys.
• Running out. Chats on the plan stop, and say when they start again. Packs of $10 of extra credits, which last a year, are on the account page; they’re spent once the month’s run out, if you turn that on there.
• Cancelling. Manage on the account page opens Stripe’s billing page, where you cancel or change the card. The plan runs to the end of the month you’ve paid for.
• Unlinking. Unlink in Models & keys, or nolune nolune-plan logout. Signing out of nolune’s API on the account page ends the link too.

API keys

nolune key set anthropic       # or openai / openrouter / xai

nolune checks a key before it stores it. OpenAI’s key also makes pictures on the Images page. Firecrawl’s key (nolune key set firecrawl) is for searching the web, which works without one up to a daily limit.

Grok, with an xAI key

With nolune key set xai (a key from console.x.ai), presets run on xAI’s Grok models: nolune preset add grok-4.7 --provider xai, or the xAI chip in Add a model, which lists the models the key can use with their context windows. Each model gets the reasoning effort it takes nearest the chat’s. Grok sees pictures (sent as JPEG or PNG); PDFs are saved for the agent and named in the message. Chats are paid from the key’s team credits; when they run out, or the team hits its spending limit, the chat says so.

Through OpenRouter

With nolune key set openrouter, a preset can run any model OpenRouter serves that can call tools: nolune preset add deepseek/deepseek-v4.1-flash --provider openrouter. OpenRouter’s ids name the model’s maker. Pictures and PDFs go only to models that take them; for the others, they’re saved for the agent and named in the message, like any other file.

On your own servers

Chats can run on model servers of your own, like Ollama or LM Studio on this computer, or vLLM on a machine with GPUs. Add each one as a custom provider, with the name you want to see, under Models & keys (Add custom provider, under the API keys) or with nolune provider add Ollama http://localhost:11434 (--key if it wants one). It speaks OpenAI’s API unless you pass --api anthropic (Ollama, LM Studio and oMLX have both; llama.cpp’s server only Anthropic’s). Then it’s a provider like the others: nolune preset add qwen3:8b --provider Ollama, or its chip in Add a model. nolune asks it for its models to check it. The model must be able to call tools; pictures and PDFs reach it as their paths.

On your own plan instead of an API key

Chats can run on a subscription someone in the family already has. Pick one in a new profile’s welcome, or with nolune <plan> setup and a preset on that provider (nolune preset add claude-opus-5-5 --provider claude-plan, or on Models & keys). nolune <plan> status says who the plan is signed in as. Plan limits assume one person’s ordinary use: keep busy automations and subagents on an API key preset. See DESIGN.md for what works differently.

• claude-plan: Claude Pro or Max. Chats run through Claude Code on this computer, unmodified, signed in to your Claude account; Claude Code keeps the sign-in and nolune never sees it. Setup installs it with Anthropic’s installer, asking first, and starts Claude Code’s own sign-in in your browser, so run it in a terminal on this computer. Anthropic counts this as Agent SDK use of your subscription.
• chatgpt-plan: ChatGPT Plus or Pro. Nothing to install: nolune uses OpenAI’s Sign in with ChatGPT for open-source apps (in preview), and its requests count toward your plan’s usage, shared with ChatGPT and Codex. Continue with ChatGPT under Models & keys, or nolune chatgpt-plan setup, opens ChatGPT’s sign-in page, where you allow nolune to use your plan. OpenAI sends the browser back to 127.0.0.1, this computer: in a browser here that’s all, and from a phone or another computer the page it ends on doesn’t load, so copy its address and paste it under Models & keys (or into setup at a terminal). nolune keeps the sign-in in ~/.nolune/chatgpt.json, readable by you only. See and limit what nolune uses in ChatGPT’s usage settings, where you can also disconnect it. nolune chatgpt-plan models lists what the plan offers (OpenAI lists a new model only to Codex versions made for it, so nolune asks for the list as Codex’s latest release, which it looks up on npm), and nolune chatgpt-plan logout signs out. Earlier versions ran Codex for this: its folder, ~/.nolune/codex, isn’t used any more and can be deleted.

Switching models in a chat

The chip in the message box picks the model and how long it thinks, in a new chat or an existing one. After a change the next reply reads the whole chat again (it isn’t cached for the new setting yet), so it’s slower and costs more once; the chat asks before that happens.

nolune preset default <name> sets the model new chats start with, and nolune preset list shows them all.