Usage and limits
Claude Code usage limits: how to check and stretch your quota
Two clocks run at once: a rolling session window and a weekly cap. Learn how to check both, what burns quota fastest, and how to make an allowance last.
Read the guideThe multi-agent workspace for Windows
Run Claude Code, Codex, Gemini and local models in one Windows workspace. See who needs you, choose Away when work can continue, and review the results together.
Product tour
Four agents, one project, one window. Pick a step to look closer.
Rebuilt in HTML from the real app. Sample project and accounts.
Works with
Claude Code, Codex, Gemini, Grok and anything else that runs in a terminal.
Claude Code and Codex: full support, including usage, model, context, history and Chat. Grok: usage and history. Gemini CLI: its own accounts. Local Ollama models: direct Chat, read-only review and Assistant with supported tools, or a terminal. Any other CLI runs in a real terminal.
Product names and logos are trademarks of their owners, shown only to identify tools Towfu can run. Their use does not mean any affiliation with or endorsement by them; Towfu is independent.
And there is more
Choose Chat, Plan, Ask first, Edit or Auto where supported. Use your subscription, an API account or a local model.
The same agent keeps the same chat on another account. A different agent gets a linked continuation and handoff notes.
Find a previous conversation and resume it when you are ready.
Sign in through the official CLI, with a separate profile for each account.
Give each agent its own folder and branch. Board tasks can start an agent in a worktree after your confirmation; review the changes before creating a draft PR.
Towfu adds no telemetry. Each CLI or cloud model connects to its own provider.
Hold to speak, see a live local transcript, then release to insert. Personal spellings live in Voice settings.
Ask what needs you across projects and the Board, or research public sources. Use a tool-capable model, including local Ollama. Assistant cannot edit files or run commands.
Enable Away globally, by project or per terminal. It takes recognised default answers and logs them; confirm before enabling, as it can approve destructive changes.
Save an agent’s account, brief and instructions. Schedule recurring work while Towfu is open; pause or resume when you choose.
Ready when you are
Windows 11, 64-bit. Bring your own coding-agent CLI and provider account.
How it works
Towfu is an independent workspace. It does not include AI model access or raise provider usage limits.
Install the coding CLIs you want to use, or connect a local model for native Chat and review.
Add one account or several. Sign in through the official CLI, add an API connection or connect a local model.
Open a project and start a session, or confirm a task from the Board.
Simple pricing
One monthly plan unlocks the whole workspace.
Your Claude, ChatGPT or other provider plan is separate and paid to them.
Your local price, with any tax, is shown at checkout.
No. Towfu runs the official CLIs you install yourself, signed in with accounts you already pay for. Your provider plan, limits and terms stay as they are.
Claude Code and Codex get full support: their own account profiles with official sign-in, five-hour and weekly usage, the model and context in each terminal, history and resume, and Chat.
Gemini CLI and Grok CLI get their own account profiles too, and sign in through their official CLIs. Grok shows its weekly limit and can resume past chats. Gemini CLI reports no usage, so Towfu shows a dash.
Installed coding CLIs such as Aider, OpenCode, Cursor, Copilot and Qwen are detected in New agent. You can also choose Other CLI in Accounts. For any other tool, open Engines & advanced settings in Accounts, choose Add your own CLI agent and point it at an .exe, an npm command, or a Node, Python or PowerShell script. It runs in a real terminal, and its sign-in, usage limits and approvals stay with the tool: Towfu doesn't claim to read them.
A plain PowerShell terminal is always there too, no account needed.
See pricing and the refund policy. If your subscription ends, nothing on your PC is deleted; Towfu just needs an active subscription to start agents again.
Towfu reads them from each CLI's own usage or status output. It never calls private provider APIs. When a reading isn't clear, Towfu shows unknown rather than guessing.
For the same agent, the same chat moves to another account. Switching to a different agent opens a linked continuation with the task, decisions and next steps from the handoff note. Another account must be available; model memory is not transferred.
After you confirm, Away automatically accepts agent approvals and answers questions with the recommended or default option. You can apply it globally, by project or to one terminal from its plane badge. Each session has an approval/held log; Settings offers an automatic off timer and turning it off when you return.
It can approve destructive commands and changes outside your project. Enable it only for work you are comfortable letting continue. Unrecognised prompts can still wait for you.
The Agents dock groups sessions by status or project, with Needs you near the top and Assistant pinned above the groups. Search or filter the list, close finished work and hide idle sessions. Failed sessions, questions waiting over 20 minutes and silent mid-turn sessions over 15 minutes can show Stuck with a reason.
Yes. Add an Ollama or LM Studio connection in Accounts to chat directly in Towfu. Tool-capable models can review project files and act as Towfu Assistant, including public web search. Local inference runs on your PC; web tools connect to the sites you request. You can also use a terminal tool such as OpenCode or Aider.
Use the Kanban Board to keep tasks, running work and review in one place. Starting a card asks for confirmation and can give its agent a separate worktree. Review the changes and create a draft pull request when ready.
Saved agents keep a reusable brief, account, project, memory and selected local instruction skills. Changes apply to new conversations. Towfu Assistant can read your Board alongside project and session status.
Yes. Use read-only project review with a supported local or API model, or choose Plan in Chat. Towfu can read the supplied project context and propose next steps without editing. Direct local/API Chat does not need another coding CLI. Modes and tools depend on model support.
When you do edit, Towfu checks unsaved file drafts before closing and keeps them if saving fails. Review model changes before approving them.
Your chosen CLI or cloud model receives the prompts and project context you give it, just as it would outside Towfu. Local models run on your PC; web searches and page reads connect to public sites. Towfu adds no telemetry. Account, voice and phone connections are described in the privacy policy.
Your next project starts here
Keep your accounts. Give your work one home.
Windows 11 · Your own provider accounts
From the Towfu blog
Usage and limits
Two clocks run at once: a rolling session window and a weekly cap. Learn how to check both, what burns quota fastest, and how to make an allowance last.
Read the guideAccounts
Give each account its own config folder and a PowerShell function to launch it. Here is what that separates, what it shares, and how to check which account is active.
Read the guideWindows setup
Native or WSL, the PowerShell installer, what Git for Windows now does, and the fixes for the errors people hit most.
Read the guide