Claude Usage Limits: 7 Habits to Stop Hitting Your Cap
Still hitting "You've hit your session limit" in the middle of a Claude Code task, then waiting hours for the window to reset? The same work can drain your allowance several times faster depending on how you carry the conversation.
A Claude usage limit is the cap on how much you can use Claude within a set period on a subscription plan. Subscription plans have a session limit that refills on a five-hour window plus a weekly limit, and both are measured in the tokens Claude processes, not the number of messages you send. This guide explains why usage climbs, which commands show where it goes, and which habits save the most, plus how daglo summaries keep long meeting transcripts out of your context.
Four reasons your Claude usage drains fast
Claude Code sends your whole conversation with every request. In a session that has been open all day, even a one-line question draws usage for everything that came before it.
- Long context: every file read and command result stays in the conversation and rides along with each later request.
- Cache expiry: on a subscription, the prompt cache (which lets Claude re-read content it has already seen at a lower cost) lasts one hour. Come back after a longer break and the first request reprocesses your full context.
- Large files and logs: asking Claude to read a several-thousand-line log puts all of it into context.
- Subagents and scheduled tasks: subagents return only a summary, but their own requests still count against your plan. Scheduled tasks such as /loop send your full context even while you are away.
Check where it goes with /usage and /context
Saving starts with knowing what costs the most. In Claude Code, /usage shows your plan usage and what consumed it, and /context shows what currently fills your context window.
- /usage: on Pro, Max, Team, and Enterprise plans it shows usage bars plus each skill's, subagent's, and MCP server's share. It also flags behaviors, such as long context or cache misses, that account for 10% or more of recent usage. Press
dorwto switch between the last 24 hours and 7 days. - /context: shows how much space the system prompt, CLAUDE.md, MCP tools, and conversation history each take.
- Keep in mind: /usage figures are estimates from this machine's history, so usage from other devices or claude.ai is not included.
Seven habits that keep you under the limit
The two habits with the biggest impact are running /clear when you switch to unrelated work and matching the model to the task. The rest keep less material from piling up in context.
| Habit | How | What it cuts |
|---|---|---|
| Clear between tasks | Name the session with /rename, run /clear, and come back with /resume | Stale context |
| Compact only when needed | Tell /compact what to keep, e.g. /compact Focus on the API changes | Long history |
| Match the model | Use Sonnet for most coding and switch to Opus with /model for hard design work | Usage per request |
| Keep CLAUDE.md short | Keep essentials under 200 lines and move workflows into skills | Fixed context every session |
| Turn off unused MCP servers | Disable servers you are not using in /mcp | Tool listings |
| Filter logs first | Use a hook or grep so only error lines reach Claude | Large files and logs |
| Be specific | Name the file and function instead of asking to "improve the codebase" | File reads |
/compact reads the whole conversation to summarize it, so it is a large request in itself. When you don't need continuity, /clear costs nothing. For moving instructions into skills, see our Claude Code skills guide, and for an open-source tool that compresses tool output, see our Headroom overview.
Feed Claude a daglo summary, not the full transcript
Pasting a full meeting transcript into an AI chat adds a large block of context from a single one-hour meeting. Keep the transcript in daglo and pass along only the summary you need, and the same question runs on a fraction of the context.
- Transcribe and summarize: upload the meeting recording to daglo to get a speaker-separated transcript and a summary.
- Ask only what you need: ask Board Chat something like "List only the decisions and action items" and get an answer grounded in the transcript.
- Pass the summary on: paste just that summary into Claude. If you need an exact quote, find that part in daglo and add only that section.
In short, your usage limit drains with conversation length and accumulated context. Check /usage to see what costs the most, then clear between tasks, match the model to the job, and keep CLAUDE.md short. Plan limits and commands change often, so check the official docs as well.
#Claude #ClaudeCode #ClaudeUsageLimit #TokenSaving #CLAUDEmd #AIProductivity #PromptCaching #daglo
Where talk becomes work.
Get started with 1,000 credits a month — no credit card required.