5 Free Repos That Make Your Claude Code Limit Last Longer

If you keep running out of Claude Code usage before lunch, five free GitHub repos help your limit last longer. Three cut what Claude reads and writes: caveman makes its replies shorter, rtk trims long command output before Claude sees it, and code-review-graph maps your project so Claude opens only the files it needs.

The other two don't save a single token. They show you where it's going: Claude Code Usage Monitor warns you before you hit the wall, and claude-usage shows which sessions burned the most.

The savings figures below are the makers' own unless we say otherwise, and yours will depend on how you work. The install lines come last, in order.

What does each repo do for your usage limit?

caveman: shorter replies, same code

caveman makes Claude talk like a caveman: the answer first, no greeting, no recap, with code and error messages left exactly as they are. Claude Code sends the whole conversation with every step, Claude's own replies included, so shorter replies make every later step smaller too.

Its full install also shrinks logs and test output before Claude reads them. Its own test across 54 Claude Code runs measured 33.2% fewer input tokens per session, and an independent test of 10 token savers measured 37 to 38%. It has over 115,000 stars on GitHub.

rtk: trim long test and build output before Claude reads it

When Claude runs your tests or a build, the whole output goes back to Claude as context. rtk compresses it first, keeping only the failures from a test run, for example. Its makers say that output gets up to 90% smaller, and they're upfront that your bill doesn't shrink by the same amount.

It pays off when Claude reads long logs. The independent test, on a task with little output, measured no saving. It has over 84,000 stars.

code-review-graph: send Claude straight to the right files

Without a map, Claude opens file after file to find the right one. code-review-graph maps your project's functions, imports and tests, and keeps the map updated as you edit, so Claude reads only the files a change touches.

Its own benchmark's median is about 63 times fewer tokens per question. That's measured against reading the whole codebase, which its makers call an upper bound no real agent pays, and the independent test measured 5% on a small project. So put it on your biggest one. It has over 32,000 stars.

Claude Code Usage Monitor: see when you'll hit the wall

Usage Monitor is a live dashboard in your terminal: progress bars for your current 5-hour session, how fast you're burning through it, and when it predicts you'll run out.

Tell it your plan, or let it work out your limit from your last eight days of use. It has over 8,700 stars, and its latest release came out in June 2026.

claude-usage: find the sessions that burn your limit

Claude Code already writes a usage log on your machine for every session. claude-usage turns those logs into charts of your tokens by model, session, project and day, so you can see which sessions burn the most and which habit to fix first.

It works on API, Pro and Max plans. Its cost figures use API prices, so on a subscription treat them as a guide, not a bill. It has over 2,200 stars.

How do you install the five repos, and in what order?

Install them in the order from the reel, the token savers first. Each one runs from a terminal.

1. caveman. This adds it as a Claude Code plugin that starts on its own in every new session. Say stop caveman to go back to normal.

claude plugin marketplace add JuliusBrussee/caveman && claude plugin install caveman@caveman

For the full install, which also shrinks logs and test output, use its installer instead. It needs Node.js 22.13 or newer and sends usage stats by default (never your prompts or code). caveman telemetry off stops them.

npx @caveman-ai/cli@latest

2. rtk. Install it with Homebrew on a Mac or Linux, or with winget install rtk-ai.rtk on Windows. Then turn on its hook for Claude Code and restart Claude Code.

brew install rtk
rtk init -g

Run rtk gain to check it worked: you should see a savings dashboard, not "command not found". It only covers commands Claude runs in the shell, not Claude Code's built-in Read and Grep tools.

3. code-review-graph. It needs Python 3.10 or newer. Run this in your biggest project, then restart Claude Code:

pip install code-review-graph
code-review-graph install
code-review-graph build

install sets up Claude Code and your other AI coding tools. After the first build, its hooks keep the map up to date as you edit.

4. Claude Code Usage Monitor. It needs Python 3.9 or newer.

pip install claude-monitor
claude-monitor --plan pro

Swap pro for max5 or max20 to match your plan, or run plain claude-monitor to let it work out your limit. Keep it open in a second terminal window. If pip refuses with an "externally-managed-environment" error, use uv tool install claude-monitor instead.

5. claude-usage. It needs Python 3.8 or newer and nothing else.

git clone https://github.com/phuryn/claude-usage
cd claude-usage
python cli.py dashboard

On a Mac or Linux, type python3 instead of python. The dashboard opens in your browser at localhost:8080, and there's a VS Code extension too.

For mods that show what each session costs inside Claude Code, see our best Claude Code mods. To find the jobs costing your team the most time, take the free AI bottleneck test, or read our insights on AI implementation.