What it is

Clauder is your own client for Claude Code. It shows everything that already sits in ~/.claude/projects/**/*.jsonl — projects, sessions, messages, the model's thinking, tool calls — and lets you start a new chat, a new project and carry the work through: from your phone, from the browser, from a window on your desktop or straight from Telegram.

The agent meanwhile runs on your machine, in your directories, with your MCP servers and your docker.

There is no storage of its own

This is the main thing to understand about how it works. Clauder reads and writes the same files the CLI does:

• a chat started in the browser shows up in claude --resume in the terminal;
• a session started in the terminal in the morning opens on your phone in the evening and continues from there;
• a chat renamed in the CLI is called the same in every client.

Next to the sessions directory there is an SQLite file (clauder.db) — but that is an index of token spend and a place for favourites, your own titles and "read" marks, not a second storage. Delete the file and it will be rebuilt by scanning.

What it takes

One requirement: a route to your machine from wherever you work. The local network, a VPN on the phone, a one-click tunnel or your own domain behind a proxy — details in Access from outside. Telegram doesn't even need that: the bot reaches out by itself, no inbound route required.

The machine, however, has to be on and online. Cloud options survive a closed laptop, Clauder does not.

What it is for

Starting work without walking up to the computer. A new chat in any directory of the machine (a folder can be created on the spot), a choice of model and effort, images from the gallery or the camera — all of it from the phone.

Seeing what the agent did. The session transcript in a readable form, not a wall of JSONL. An open session updates live — you watch a reply that was started in a terminal in the next window.

Answering without going back to the terminal. If Claude asked a question (AskUserQuestion), you can answer it from Clauder — even when the session is run not by Clauder but by claude in the terminal or in VS Code. That needs hooks.

Seeing who is working right now. A separate screen with live chats, their subagents and background tasks: what each one is busy with and how long it has been going.

Running it where you need to. A session doesn't have to run on the server: its place can be a folder on your own disk handed to the browser, another machine through the worker bridge, or the phone itself — the choice is the first question of a new chat (Where the session runs).

Handing over passwords without revealing them. The agent asks for a password or a token as a card, the server substitutes the value, and only the secret's name stays in the conversation (Secrets).

Keeping everything in one list. Sessions from WSL and from Windows sit side by side even though they live in different profiles. Plus search across the whole conversation history of every project at once.

How this differs from the official ways

Claude Code has its own ways to reach the agent from afar: Remote Control (claude remote-control — the session is visible on claude.ai/code and in the Claude app) and Claude Code on the web (the task runs in Anthropic's cloud VM on a GitHub repository, or on a bundle of your local one uploaded there). Clauder solves the same problem from the other end: it doesn't attach one session to someone else's cloud, it shows and drives everything that is already on your machine.

	Clauder	Remote Control	Claude Code on the web
Where the agent runs	your machine	your machine	Anthropic's cloud VM
Which sessions you see	everything in ~/.claude — past, foreign, both sides of WSL/Windows	only the ones you enabled; after server mode stops they come back for about four more hours	cloud ones only
Archive and search	the whole archive, full-text search over conversations	no	a list of cloud sessions
A new chat from the phone	in any directory of the machine	in the directory where server mode is up, or in its worktree	in a GitHub repository, or in a bundle of your local one
Where the transcript goes	your own JSONL	JSONL on your side + a copy on Anthropic's servers while connected	the cloud
Notifications	Web Push, Android, tray, Telegram — about everything on the machine	pushes for your own sessions	same place
Spend and limits	a screen of its own: the window, weekly limits, pace, breakdowns	/usage as text	/usage as text
Price	your server, MIT	included in the subscription	included in the subscription

In short: the official ways give you a window onto one session you opened yourself; Clauder gives you a remote for the whole machine — lists of projects and chats, the archive, search, files, spend, foreign sessions and the ability to start new ones.

Where the official way is better

This is an honest list, and it matters more than the sales pitch:

• Nothing to host and nothing to keep switched on.
• The cloud runs tasks in batches and in parallel without eating your CPU.
• Tool approvals there are part of the product. Clauder has them too, but the default is bypassPermissions (see Security).
• It is a supported product with guarantees, while Clauder is a home tool.

What next

• Installation — the app or docker compose.
• First run — the key, the phone, the first chat.
• Clients — how the five ways to connect differ.