Ollama Sidekick

Ollama Sidekick brings local and user-configured AI into Chrome's side panel. Chat with models running on your own computer, connect OpenAI-compatible APIs you choose, and ask questions about the page you're viewing without leaving the browser. LOCAL-FIRST AI IN YOUR BROWSER - Chat with local Ollama models without sending requests to a third-party AI service. - Add multiple OpenAI-compatible cloud connections when you want to use hosted models. - Switch between local and cloud models from one model picker. - Stream responses, cancel in-progress requests, and see connection diagnostics when something needs attention. ASK ABOUT THE PAGE IN FRONT OF YOU Page context is optional and controlled by you. Choose the amount of context to send: - Whole page - Highlighted selection - Headings and page outline - Code blocks - A custom CSS selector Pin page context to a chat, re-pin it when the page changes, and use page-type-aware quick actions for articles, documentation, products, GitHub pages, and video pages. You can also attach an opt-in screenshot when the selected model supports vision. USE YOUR OWN MODEL CONNECTIONS - Connect Ollama or any OpenAI-compatible API endpoint you explicitly configure. - Discover models through a provider's `/models` endpoint or enter model IDs manually. - Set or inspect context-window and maximum-output limits per provider or model. - Keep long conversations usable with context-capacity indicators, rolling summaries, and bounded request budgets. OPTIONAL AGENT MODE Agent Mode is an opt-in experimental feature for policy-controlled browser tasks. It can observe authorized pages, navigate tabs, interact with page elements, and report verified results from the side panel. - Exact site access and account context are bound to each task. - The agent pauses for passwords, one-time codes, payment details, external communication, purchases, deletion, permission changes, and other consequential actions. - Each action is checked against the task policy and its result is independently verified when possible. - Tasks are bounded by step, model, token, screenshot, and time budgets, with recovery and restart safeguards. - Agent Mode does not continue after the side panel closes and cannot start autonomous website downloads. SEARCH, MEMORY, AND CHAT ORGANIZATION - Brain mode searches pages you choose to index locally and adds relevant passages to a request. - Use multiple chats, search chat history, and keep drafts when switching conversations. - Export a chat as Markdown or export all chats as JSON/Markdown. - Import a validated JSON backup. - Keep model-generated task memory inspectable, correctable, deletable, or disabled from the active task UI. PRIVACY BY DESIGN - No account, sign-up, analytics, advertising, or developer-owned data server. - Chats, settings, indexes, summaries, and Agent Mode records stay in local browser storage. - Cloud requests go directly to the provider URL you configure; the extension developer does not proxy them. - Page context is sent only when you enable it for a request, and screenshots are sent only when you opt in. - API keys are stored in dedicated local extension storage, are not synced, and are not exposed to webpages. - You can use local Ollama when you want the model request to stay on your own machine. REQUIREMENTS For local use, install and run Ollama: 1. Install Ollama from ollama.com 2. Download a model: ollama pull llama3.2 3. Start Ollama with CORS enabled for the extension: - macOS/Linux: OLLAMA_ORIGINS='chrome-extension://*' ollama serve - Windows (Command Prompt): set OLLAMA_ORIGINS=chrome-extension://* && ollama serve - Windows (PowerShell): $env:OLLAMA_ORIGINS='chrome-extension://*'; ollama serve Note: If `chrome-extension://*` does not work, use `OLLAMA_ORIGINS='*'` but be aware this allows any website to access your local Ollama API. 4. Click the extension icon to open the side panel Alternatively, add an OpenAI-compatible HTTPS API base URL, API key, and model IDs in Diagnostics & Settings. Cloud use does not require Ollama. SUPPORTED MODELS Works with models installed in Ollama and models exposed through an OpenAI-compatible API. Vision features require a model that accepts images. Brain mode and Agent Mode use the selected model and respect its configured or discovered capacity. TROUBLESHOOTING If you see a connection error: 1. Make sure Ollama is installed (download from ollama.com) 2. Stop any existing Ollama process: - macOS/Linux: pkill ollama - Windows: Close Ollama from the system tray or use Task Manager 3. Start Ollama with CORS enabled for the extension: - macOS/Linux: OLLAMA_ORIGINS='chrome-extension://*' ollama serve - Windows (Command Prompt): set OLLAMA_ORIGINS=chrome-extension://* && ollama serve - Windows (PowerShell): $env:OLLAMA_ORIGINS='chrome-extension://*'; ollama serve Note: If `chrome-extension://*` does not work, use `OLLAMA_ORIGINS='*'` but be aware this allows any website to access your local Ollama API. 4. Click the retry button in the extension Windows users: If you installed Ollama as a desktop app, you may need to set the OLLAMA_ORIGINS environment variable in System Settings > Environment Variables, then restart the Ollama app.