Save on your AI agent's usage so the limit lasts all week

Claude counts more than messages. Anthropic lists conversation length, the size of attachments, the model and effort level you choose, and tools such as web search and research. A long conversation costs more because every new message makes Claude read the whole conversation again.

In Claude, open Settings and then Usage. It shows how much time is left in the current session and when your weekly limit resets. Paid plans have a five-hour session limit and a weekly limit, shared between the web, the desktop app, the phone and Claude Code. In Claude Code, type /usage.

Take them in this order. The first four cost nothing and change the most.

Three instructions you can copy. The first goes under your project's instructions. The other two you paste into a conversation when you need them.

Claude Code has its own commands for this. These come from Anthropic's page on managing costs.

• Type /usage to see what counts against your plan and what to change.
• Type /clear when you switch to unrelated work. Use /rename first if you want to come back to the session.
• Type /compact followed by what to keep, for example: /compact keep the decisions and the open questions. Use it when a long task has natural phases.
• Use /model to pick a lighter model for most tasks, and /effort to lower the effort on simple ones.
• Type /context to see what takes space, and /mcp to switch off connections you are not using.
• Keep CLAUDE.md short, under 200 lines according to Anthropic, and move special instructions into skills, which load only when needed.
• Use plan mode (Shift+Tab) for big tasks, press Escape early if it heads the wrong way, and give it a way to check its own work, such as a test.
• Let a subagent read long logs or documents, so only a summary comes back to your main conversation.

A session left open for hours can use far more than your activity suggests. These are the usual reasons.

• Claude Code sends the whole conversation with every request, so a one-line question in a session that has been open all day still draws on all of it.
• The first message after a long break reprocesses everything, because the saved copy has expired.
• Scheduled tasks and loops run on their interval even while you are away, and each run sends the whole context.
• Subagents and agent teams send their own requests on top of yours.
• Compacting a large conversation is itself a large request. When you do not need the thread, start fresh with /clear instead.

Read it once, then keep it near the screen for a week.

• Did I put everything related in one message?
• Did I give the goal, the material and the result I want?
• Is this a new topic? Then start a new conversation.
• Does the material I reuse sit in the project?
• Do I really need to attach the whole file?
• Does this task need the strongest model, or a web search?

When you spend less on the small things, the limit stops deciding when you work.

• A weekly rhythm: the heavy work right after a reset, the light work at the end of the window.
• A project per kind of work, each with its own short instruction and the material it reuses.
• A habit of switching conversation at every new topic, with a five-line summary as the bridge.