LastBrowser launches open-source browser with Sidekick

LastBrowser has launched an open-source browser for Windows and Linux with a built-in AI assistant called Sidekick. The software is available free for private and commercial use under the GPL-3.0 licence.

Developed in Innsbruck, Austria, by Founder Dominik Rainer as an independent project, the browser enters a crowded market with a product aimed at users who want AI tools inside the browser window while keeping tighter control over what the assistant can access.

Sidekick sits in a split view beside the active web page. By default, it can access only selected text, the page title and the URL. Full page text is available to the assistant only when the user explicitly allows it for the current Space.

That permission is temporary and ends after a short period or when the user leaves the page. The approach is intended to limit exposure of information when users are working with internal systems, customer records or confidential documents.

"Most people spend their working day inside a browser, yet AI help still lives in a separate tab you have to feed by hand," said Dominik Rainer, Founder of LastBrowser.

"I wanted an assistant that sits right next to the page and only sees what it needs to see. Making LastBrowser open source under GPL-3.0 means anyone can check exactly how that works," he said.

Spaces and access

LastBrowser groups browsing activity into Spaces, which separate cookies, caches and logins. That allows users to run work and personal accounts for the same online service in one browser window without mixing sessions.

Permissions for Sidekick follow those boundaries. A page made available to the assistant in one Space is not automatically exposed in another.

The browser also includes a vertical tab sidebar for users managing large numbers of open tabs. It also has a built-in terminal in the sidebar that uses PowerShell or CMD on Windows.

Sidekick can suggest terminal actions, but each one requires manual approval by default. Potentially dangerous commands are flagged before execution.

Open model

Users can run the AI assistant through their own API keys with cloud AI services or locally on their own machines with local models, including through Ollama. The browser does not require registration, a point the developer has presented as part of its privacy position.

The source code has been published publicly under GPL-3.0, and development is taking place in the open, with release notes and issue tracking available on GitHub. Rainer said the project had initially considered a commercial licence before opting for an open-source model.

LastBrowser already has a few hundred users, according to the company. It also offers stable and beta update channels through public GitHub releases, allowing users to choose between more established builds and earlier access to new features.

Developer focus

The browser includes what it calls an Extension & MCP Hub, which lets users connect browser extensions and Model Context Protocol tools. It also supports Widevine DRM playback for streaming services and includes a Privacy Shield feature.

On Windows, both the installer and portable builds are code-signed. Linux support is currently described as a preview.

The launch reflects a broader shift in browser design as developers test ways to embed generative AI directly into the browsing experience rather than keeping it in separate chat interfaces. At the same time, those efforts have drawn scrutiny over how much page content assistants can view and how user consent is handled.

Rainer positioned the browser as an alternative to products that gather more user data. "We are not trying to build yet another browser that collects as much data as possible," he said.

"LastBrowser is meant to be a tool that works for its user: transparent code, local options for AI and clear limits on what the assistant can access," he said.