PrivateGPT Sidebar

Chat with local AI models in your browser's side panel. Connects to Ollama on your computer, so your data stays private.

PrivateGPT Sidebar — Chat With Your Own Local AI, Right Beside Any Page PrivateGPT Sidebar puts a full AI chat assistant into your browser's side panel and connects it to Ollama running on your own computer. Your prompts and your model's answers never leave your machine. There is no cloud service in the middle, no account to create, no subscription, no API key, and no usage meter counting your tokens. If you have ever wanted the convenience of a browser AI assistant without handing your questions, your drafts, your code, or your client data to a third party, this is that assistant. WHY A SIDEBAR Popup windows disappear the moment you click away. New tabs bury the page you were actually reading. PrivateGPT Sidebar lives in Chrome's native side panel, so your conversation sits next to your work instead of on top of it. Read a document on the left, ask questions on the right. Compare a spec against your notes. Draft a reply while the original message stays visible. The panel stays open as you move between tabs, so your train of thought stays intact. COMPLETELY PRIVATE BY DESIGN Privacy here is not a policy promise, it is the architecture. The extension talks to a single address: the Ollama server listening on port 11434 on your own machine. There are no analytics scripts, no telemetry pings, no remote logging, and no hidden background uploads. The extension declares only two permissions — declarativeNetRequestWithHostAccess and sidePanel — and out of the box its network access reaches nothing but your own machine; any other Ollama host is granted by you, at the moment you type it in. Because everything runs locally, it also works offline. On a plane, on a train, on a locked-down network, behind a corporate firewall: if Ollama is running, your assistant is running. That makes it a realistic option for work that genuinely cannot go to a public AI service — medical and legal drafting, unreleased source code, internal financial models, HR material, student assessment, journalism with sources to protect, and anything covered by GDPR, HIPAA, or an employer's data policy. USE ANY MODEL YOU HAVE INSTALLED The extension reads your installed model list directly from Ollama and offers it in a dropdown. Llama, uncensored community builds, your own fine-tunes — if you can pull it with Ollama, you can chat with it here. Switch models mid-session to match the job: a small fast model for quick rewrites, a larger reasoning model for hard problems, a coding model for debugging. Your chosen model is remembered so you pick up where you left off. FEATURES THAT MAKE IT USABLE ALL DAY Streaming replies. Text appears word by word as the model generates it, so you can start reading immediately and stop early when you already have your answer. Stop button. Interrupt a long generation instantly instead of waiting it out or reloading. Real Markdown rendering. Headings, lists, tables, links, and fenced code blocks render properly, and everything is sanitized before it is displayed, so a model's output can never inject anything into the page. One-click copy. Copy an entire reply to the clipboard without selecting a single character. Conversation memory. The extension carries conversation context between turns, so follow-up questions like "now make it shorter" or "explain that last part" work the way you expect. Named, saved chats. Give a conversation a name and keep it. Reopen it later from the History dropdown with its full transcript and its context restored, so you can resume a thread days later. Delete the ones you no longer need. Saved chats stay in your browser's local storage — again, on your machine only. Smart auto-scroll. The view follows the response as it streams, but it steps out of the way the moment you scroll up to reread something, and resumes when you scroll back down. Light and dark themes. A one-click toggle that also follows your system preference by default and remembers your choice. Keyboard-first input. Ctrl+Enter (or Cmd+Enter) sends. The input box grows as you type and the panel adjusts around it. Remote and network hosts. Running Ollama on a home server, a workstation down the hall, or a GPU box in the next room? Point the hostname field at that machine and approve the one-time Chrome permission prompt for that address; from there the extension handles the origin headers for you, so you can use serious hardware from a lightweight laptop. Nothing outside your own machine is reachable until you explicitly allow it. AVAILABLE IN 40+ LANGUAGES The interface is fully localized, including Arabic, Chinese (Simplified and Traditional), Japanese, Korean, Hindi, Russian, German, French, Spanish, Portuguese, Italian, Dutch, Polish, Turkish, Persian, Hebrew, Thai, Vietnamese, Indonesian, and the Nordic, Baltic, and Central European languages. GETTING STARTED Install Ollama from ollama.com, pull at least one model, make sure it is running, then open the side panel. Your models appear automatically and you can start typing. There is nothing to sign up for and nothing to configure beyond that. Fast, private, offline-capable AI in your browser — on your terms, on your hardware.