Claude Ease

Claude Ease is a companion for the Claude Code CLI that lives inside VS Code. It does two things the raw CLI and the stock chat panel leave to you: it teaches Claude your whole codebase up front, and it handles the fiddly parts of a session — which model to use, how close you are to your limit, and what to do next — so you can stay on the actual work.

It runs entirely through your own local claude CLI, so it works with whatever plan you already have, Pro/Max subscription or pay-as-you-go API.

Built by Atish Paul.

Requirements

• The Claude Code CLI, installed and signed in (claude on your PATH).
• A folder open in VS Code.

What it does

Understands a project in one step

Click Understand Project once. Claude Ease reads the repository, strips every source file down to its structure — imports, exports, classes, function signatures — and writes a compact map. A 500 KB codebase usually comes out around 7,000 to 20,000 tokens, roughly 70 to 90 percent smaller than the source. From that map it builds an architecture diagram and a short, written memory of the project: what it is, its modules, data model, the main runtime flows, conventions, and the gotchas worth knowing.

That memory is then loaded into the first message of every new chat automatically. Claude starts each conversation already knowing the project, instead of rediscovering it file by file every time.

Knows which folders matter

A lot of people open one parent folder that contains several projects at once — a backend, a web frontend, an admin panel, a mobile app. Claude Ease detects the separate folders, lets you pick which ones to include the first time you build the map, and labels each one in the memory (which folder is the backend, which is the frontend, and so on). You set the scope once instead of explaining your layout in every prompt.

Picks the model for you

Most people do not want to think about model selection on every message. In Auto mode, a quick, cheap classifier looks at each request and routes it to the least expensive model that can handle it: Haiku for simple questions and small edits, Sonnet for normal feature work and debugging, Opus for hard problems like architecture or tricky algorithms. You can still pin a specific model whenever you want. Models that are unavailable or suspended on your account are detected and marked.

Shows your usage honestly

A strip above the input shows your usage from the moment you open the chat — you do not have to send a message first. It displays the exact time your rolling five-hour window resets with a live countdown, how many tokens you have used in the current window, and an estimate of what your next message will send before you type it. On a Pro or Max plan the dollar figures are clearly labelled as an API-equivalent reference rather than money you are spending, because a subscription is a flat fee.

Session warm-up — schedule your window around your day

Claude's usage limit is a rolling five-hour window that starts the moment you first use it. If you first run something at 9:00, your windows are 9:00–14:00 and 14:00–19:00 — two windows across a 9-to-6 day. Claude Ease can do better: tell it your working hours with the inline time pickers, and it will schedule a tiny "warm-up" call a few hours before you start (around 6:00 for a 9:00 start). That shifts your window boundaries to 6:00–11:00, 11:00–16:00, 16:00–21:00, so you now touch three windows during work instead of two — meaningfully more total quota for the hours you actually code. It is fully optional, runs only while VS Code is open, and you can also trigger it manually with "Warm up now".

Urgent mode — when you just need it fixed

One tap on the Urgent button switches the whole turn into a no-friction, get-it-done mode. The assistant skips the preamble, analysis, options and summaries, makes the smallest change that fixes the problem, applies it immediately, and confirms in a single line. It never stops to ask clarifying questions — it makes a sensible assumption and proceeds. Use it for the quick fixes where speed matters more than a thorough explanation; switch it back off for normal, considered work.

A chat built for real work

• Replies stream in, with the model's thinking shown in a collapsible block you can turn off.
• When Claude runs a command you get a chip you can expand to see the exact input and the output it produced.
• When Claude edits files, the end of the turn shows the changed files — click to open or view the diff — along with one-click Commit and Run-and-verify, and a single recommended next step. These actions only show up when there were real edits or a genuine to-do list, not after a plain answer.
• You can see and kill any process the CLI leaves running, so a forgotten dev server never lingers in the background.
• Dictate with the microphone, attach files, paste an image straight from the clipboard, reference a file and line with @path:line, open the chat to the side so the file tree stays visible, and copy any code block or whole answer with one button.
• Response modes — Quick, Balanced, Deep — trade speed against thoroughness (Quick lowers the thinking budget and keeps answers terse; Deep raises it and covers edge cases). A Strict mode makes Claude push back and give blunt, critical answers instead of agreeing by default. (Urgent mode is described in its own section above.)
• The Improve Prompt button rewrites a rough request into a clear, project-aware one. It always uses the cheap model, never your expensive one.
• If the CLI ever stalls, Claude Ease checks the Anthropic status page and tells you whether there is an active incident, so you are not left guessing.

How it compares to the standard Claude extension

The official Claude Code experience is a strong chat over the CLI. Claude Ease keeps that and adds the layer around it that the stock panel does not have:

• Project memory. The standard panel reads files on demand each session. Claude Ease pre-builds a compact map and a written memory once, then injects it into every new chat, so less context is spent re-exploring and large repos fit far more easily.
• Multi-project awareness. It detects and labels separate sub-projects in one workspace and lets you scope them, instead of you describing the layout each time.
• Automatic model routing. You do not choose a model per task; the cheapest one that fits is chosen for you, which saves tokens on the many small requests.
• Real usage visibility. Exact reset time, current window usage, and a pre-send estimate are shown continuously, with subscription-aware framing — not just a number after the fact.
• Session warm-up to align your five-hour limit windows with your working hours.
• Practical session tools the stock panel does not surface: expandable command input and output, a changed-files panel with diff and commit actions, process visibility with a kill switch, status-page checks when the CLI goes quiet, prompt rewriting, speech input, and image paste.

If the built-in panel is the chat, Claude Ease is the workbench around it.

Commands

• Claude Ease: Understand Project — build the map, diagram, and memory.
• Claude Ease: Open Project Brain — the rendered memory and architecture diagram.
• Claude Ease: Choose Project Folders — scope which folders the map and memory cover.
• Claude Ease: Open Chat to the Side — run the chat in the editor area.
• Claude Ease: Minify Project — rebuild only the map.
• Claude Ease: Add File / Selection to Chat — insert an @path reference.

Where things are stored

The map and memory are written to a .claude-ease/ folder in your project. Add it to .gitignore, or commit memory.md if you want to share the project memory with your team. Everything runs through your local CLI in your workspace.

Support

If Claude Ease saves you time, you can support its development at buymeacoffee.com/atishpaul.

MIT licensed. Built by Atish Paul.