Accounts and models

Work and personal, or one plan per client: an agent signed in once has no idea which project should use which. In Wanigan an account is a folder your agent keeps its sign-in in, and each project chooses its own.

Accounts live in Settings › Accounts (also in the palette and the Go menu). Each row says who the account is signed in as, according to the agent itself, with its plan, what its limits have left and which projects use it. Wanigan never holds, reads or copies a password or a token.

1. 1

   Open the project’s settings

   In the project, click its name in the project bar and choose Project settings….

2. 2

   Pick an account for each agent

   Under Accounts, choose a Claude Code and a Codex account. Your default follows whichever account is the default. The choice applies at once, to the project’s new sessions.

   

3. 3

   Start a session

   New session (On a Mac, ⌘T·; on Windows and Linux, Ctrl+T) starts on the project’s account, marked “this project’s”, and shows what that account has left. You can still pick another for this one session.

   

Your agents’ usual folders, ~/.claude and ~/.codex, are found by themselves, and so are other account folders in your home named like them. To add another:

1. 1

   Add account

   In Settings › Accounts, click Add account beside Claude Code or Codex, give it a name you will recognise, such as “Work” or “Client: Northstar”, and choose Add and sign in.

2. 2

   Sign in in the terminal

   Wanigan makes a new folder in your home for it and opens the agent there, in a terminal inside the dialog. Follow its prompts (in Claude Code you can type /login); your browser opens to finish signing in.

3. 3

   Done

   Click Done, and Wanigan asks the agent again who it is signed in as.

Sign in on a signed-out row does the same for a folder you already have. Make default makes an account the one every project that has not chosen uses. A row’s More menu has Rename, and Remove from Wanigan, which stops offering the account and leaves its folder and its sign-in exactly as they were. Check again asks every agent who it is signed in as, and reads their limits again.

Each signed-in account shows its limits as the agent reports them: a Session and a Week meter, how much is used (or “used up”), when each resets, and how long ago it was read. New session shows the same for the account you pick, as what is left, so a session starts where there is room. There are no dollars anywhere: a plan is not billed per token.

What happens when an account runs out, and carrying a Claude Code conversation on under another account, is on Agents and the board.

Gemini CLI has one sign-in, kept where Gemini keeps it, so it has no rows in Accounts and every project uses it. Gemini takes hooks only from its own settings, so Wanigan keeps a Gemini home of its own in its data folder (GEMINI_CLI_HOME), and your ~/.gemini is never written. As each session starts, your chosen sign-in method, your trusted folders and the MCP servers that hold nothing that could be a credential are copied in, and your own skills are linked so Gemini reads them where they are.

• Gemini’s first screen may ask you to sign in or to trust the folder, so send the first message in its terminal; the message box works from the second.
• When Gemini reaches its usage limit it draws a dialog (keep trying, switch model, or stop). Wanigan shows the session as limited, and Needs you sends you to its terminal with Answer in terminal.
• Your own Gemini extensions and global GEMINI.md are not loaded in Wanigan’s Gemini home.

Some work should not leave your Mac, and some days a plan’s limit runs out. Claude Code can run on a model that lives on your Mac, through LM Studio: prompts and code go to it, not to a model provider, and no plan’s limits apply. It is optional. Without LM Studio, Wanigan works exactly as it does, and nothing is downloaded until you choose.

1. 1

   Get LM Studio

   Install LM Studio and open it once; Wanigan finds it then. Settings › Local models shows whether it is installed and whether its server is running.

2. 2

   Get a model

   Under Models Wanigan knows, Qwen3-Coder 30B runs with Claude Code. Get (17.2 GB) asks first, saying what LM Studio downloads and where (~/.lmstudio/models); Download 17.2 GB starts it. Progress shows as it goes, Stop stops it, and getting it again carries on where it stopped. Once it is there, the row says On this computer.

   

3. 3

   Start a session on it

   In New session (On a Mac, ⌘T·; on Windows and Linux, Ctrl+T), choose Claude Code, then the model under On this computer in the Model list. The first start loads it, which can take a minute. A local model takes no effort setting.

   Picture: to come

   A screenshot of the real app, with the demo’s data, of what the title above says.

4. 4

   Make it a project’s default

   In Project settings…, Local model › Start new sessions on makes the project’s new sessions start on it, and on its agent. The section appears once a local model is on your Mac.

If Ollama is running, or NVIDIA PAIR, which answers at Ollama’s address, its models are offered too, under On this computer, with Claude Code or Codex, and listed in Settings › Local models under Other models on this computer. Wanigan never downloads through them: add models with Ollama itself. A model they do not have refuses to start.