Differences from the desktop app

The Docker image runs the same app as the desktop: the same workspaces, sources, chat, agent, Studio, plugins and models. What differs comes from running in a browser on another machine.

From the desktop appCopy Anchor Link

Local models run on the processorCopy Anchor Link

The image has no GPU support yet. llama.cpp, image generation and voice run on the CPU only. Chat with a local model and Studio images are much slower than on a desktop with a graphics card. A remote connection is unaffected. The image runs on x86-64 (amd64) only.

FilesCopy Anchor Link

UpdatesCopy Anchor Link

There is no in-app updater and no update check. You update by pulling a newer image. See Updating.

Sign-inCopy Anchor Link

The desktop app has no password. The image asks for one, managed in Settings › Access, which exists only in Docker. See Access.

PrivacyCopy Anchor Link

The image keeps no usage statistics, and the privacy step of onboarding is not shown. See Privacy.

SecretsCopy Anchor Link

On the desktop, stored provider keys are encrypted with a secret in your system keychain. In Docker the secret is SURFSENSE_SECRET, or a file the image creates in the data volume.

DictationCopy Anchor Link

Browsers allow the microphone only over HTTPS or on the machine running SurfSense. From another device over plain HTTP, the dictation strip says: "Your browser allows the microphone only over HTTPS, or on the computer running SurfSense." Turn on HTTPS to dictate from other devices.

ChatGPT and plugin sign-insCopy Anchor Link

Plugin sign-ins return to the device you started them on. A ChatGPT subscription sign-in from another device ends on an error page; paste that tab's address into the sign-in dialog to finish. See Sign-ins from another device.

Claude planCopy Anchor Link

As on the desktop, a Claude plan uses your own Claude Code. In Docker you install it inside the container. The dialog shows each command with a copy button. Run them on the Docker host, in the folder with docker-compose.yml:

docker compose exec surfsense bash -c 'set -o pipefail; curl -fsSL https://claude.ai/install.sh | bash'
docker compose exec -it surfsense claude

In the second, type /login and sign in with your Claude account. Claude Code and its sign-in are kept in the data volume, so they survive image updates.

The interfaceCopy Anchor Link

• The right panel's toggle sits on the Studio heading's line, or on the chat header while the panel is hidden. The desktop has it in the title bar.
• Match system follows the browser's languages, and the theme follows the browser.
• About shows the image's version. The session log and the crash notice are desktop-only.

LicenseCopy Anchor Link

Your license works the same way. Enter it in the app; it is checked offline and isn't tied to a machine, so one license works on a desktop and a server. See License.

From the legacy Docker stackCopy Anchor Link

The image replaces the old self-hosted stack, which now lives on the legacy branch (Legacy Docker installation). They share nothing.

Both use a surfsense folder by default, so install the new image from another directory. The ports differ, so both can run side by side while you move your data.

Moving your data from the legacy stackCopy Anchor Link

There is no database migration from the legacy stack. Documents move over as markdown; chats, original files and generated artifacts don't.

1. In the legacy app, open each workspace's settings and use Export knowledge base. You get a ZIP of markdown files in the workspace's folder structure.
2. Unzip it.
3. In the Docker image, create a workspace, then in its sources choose Upload folder… and pick the unzipped folder. SurfSense indexes the files as new sources.