Does Claude Run Locally? Apps, Models and Offline Limits

Claude models run on Anthropic's servers. When you send a message, it travels to Anthropic's infrastructure, the model generates a response, and the response comes back to you. No part of the Claude model itself runs on your machine, regardless of which app you use to access it.

That said, several Claude client apps install on your computer and work directly with your local files. Claude Desktop, Claude Code and the Claude mobile apps are all local apps that send requests to the API. Your project context — the task description, the decisions and where things stand — can live in a tool like fenbs so it is available to any Claude client without being tied to one machine.

The difference between a local app and local inference

Local inference means the AI model runs on your hardware — your CPU or GPU processes the requests. This is how open-source models like Llama and Mistral can run on a developer's laptop with the right software. Claude does not support local inference. Anthropic has not released the model weights; there is no supported way to run Claude on your own hardware.

A local app means software that installs on your machine and runs there, even if it connects to a remote service for its AI responses. Claude Desktop is a native macOS and Windows application. Claude Code is a command-line tool you install in your terminal. Both are local apps in this sense, but neither runs the model locally.

Claude Desktop: what it installs and does

Claude Desktop is a native application for macOS and Windows. It gives you a desktop window for Claude conversations, system notifications and a persistent interface that does not require a browser tab. Conversations still go through Anthropic's API.

Claude Desktop supports MCP connections, which lets you connect it to local tools — files, databases and applications on your machine — through the MCP protocol. This is different from running the model locally; it means the model in the cloud can read and act on local resources you explicitly share with it.

For a broader look at what Claude Desktop does and when it makes sense, see what is Claude AI.

Claude Code: a local CLI that reads your files

Claude Code is a command-line tool you install with a package manager (npm or Homebrew). It runs in your terminal, reads the files in your project directory and sends that code to the Anthropic API for analysis. The code leaves your machine to generate the response.

Claude Code is often described as running locally because it reads your local files and edits them directly, which feels like a local operation. But the AI processing happens on Anthropic's servers. The distinction matters for anyone with strict data residency requirements. See what is Claude Code for a full picture of what it does.

For running multiple Claude Code tasks at the same time, see claude-code-run-tasks-in-parallel. Each parallel task still sends requests to the API, so usage adds up across simultaneous sessions.

Does Claude work offline

No. Claude requires an internet connection to function. Because the model runs on Anthropic's servers, every interaction needs a connection to reach them. Claude Desktop, Claude Code and the mobile apps all show an error or wait state if there is no internet connection.

If you need an AI assistant that runs without a connection — on an airplane, in a facility without internet access, or in an air-gapped environment — you would need a locally running open-source model, not Claude. There is currently no offline Claude option.

Data and privacy: what leaves your machine

With Claude Desktop, the text you type and any files you share in conversations are sent to Anthropic's API. With Claude Code, the code files you include in the session are sent. With the web app, the content of your messages and any uploads are sent.

For Anthropic's current data handling policies — how long inputs are retained, whether they are used for training, and enterprise data agreements — check the Anthropic privacy policy↗ (opens in a new tab) directly. Privacy terms change, and the current policy is the authoritative source.

If your organization has compliance requirements around where data is processed, confirm those requirements with Anthropic before using Claude Code or Claude Desktop in your environment. Enterprise agreements may include specific data handling commitments that the standard terms do not.

How to do this with fenbs

1. Create a free account and add your project to fenbs.
2. Write the task description, Decisions and rules and Where we left off once. This context is available to any connected Claude client.
3. Connect Claude Desktop or Claude Code to fenbs over MCP from Integrations.
4. At the start of each session, ask Claude to read the project context. It gets the current state regardless of which client you are using.
5. If you switch between Claude Desktop, Claude Code or the web app, the fenbs project context is the same across all three.

For Claude Cowork, Anthropic's team product, see what is Claude Cowork. For the Claude interface and how it is organized, see claude design.