Ollama Client - Local AI Chat for Ollama, LM Studio, llama.cpp

Ollama Client - Local AI Chat for Ollama, LM Studio, llama.cpp Ollama Client is a privacy-focused browser extension for chatting with local and user-configured AI providers. Connect directly to your preferred model server, manage conversations, use local knowledge, and access AI tools from the browser side panel. An optional, experimental browser agent can also try simple tasks in your browser when you ask it to. Verified Built-In Providers • Ollama • LM Studio • llama.cpp Beta Custom Providers • OpenAI-compatible servers and services, including vLLM, LocalAI, KoboldCPP, and OpenRouter • Anthropic through the native Claude Messages API • Other compatible local, LAN, or remote endpoints Features • Connect and manage multiple AI providers • Discover available models and configure custom model IDs • Switch models and monitor provider connection status • Stream chat responses with stop, retry, edit, regenerate, and fork controls • View reasoning and thinking traces from supported models • Use native and prompt-based tool calling • Search the web through your configured SearXNG, Brave, or Tavily provider • Attach local files and images to conversations • Build reusable local knowledge from uploaded documents, with on-device embeddings • Add optional webpage and browser-tab context • Use selected-text actions from the right-click context menu • Save memories and reusable context • Organise conversations with search, tags, pinning, and session management • Create custom prompt templates and system prompts • Adjust model parameters and capability overrides • Export, import, and reset locally stored data • Access privacy, permission, storage, and diagnostic controls • Use the browser side panel for a focused desktop workflow • Multi-language interface: English, German, Spanish, French, Hindi, Italian, Japanese, Russian, and Simplified Chinese Browser Agent (Experimental, Chrome only) The browser agent is an early preview and is off by default. When you turn it on in Settings and ask for something in chat, the AI can try to carry out simple tasks in your browser, such as opening a page, clicking a link or button, reading information from a page, or filling in a few fields. Longer tasks, complex sites, and sites with logins or CAPTCHAs often do not work yet. • It runs only when you ask, shows every step in the side panel, and may stop to ask you a question • You can pause or stop it at any time • Submitting forms, ordering, paying, sending and deleting ask for your approval before they happen • Passwords, payment details, one-time codes, CAPTCHAs and file uploads are never handled by the agent; it hands those steps back to you • Text on web pages cannot approve actions or change what you allowed • It may report a task as done when it is not, so check the result yourself • Results depend heavily on the model; capable hosted models work best, and small local models usually cannot complete tasks Privacy • Chats, session history, settings, knowledge, and vector embeddings are stored locally in your browser • Provider API keys are stored only on your device and excluded from backups • Requests are sent directly to the provider you configure; Ollama Client does not operate an intermediary inference server • Local providers can keep inference entirely on your device or local network • Remote providers receive content only when you choose to use them • Page, tab, file, and browser context is shared with the selected model only when you include or request that context • When the browser agent runs, page content and, for vision models, screenshots of the visible page are sent to the provider you selected so it can decide the next step; password and payment fields are masked, and screenshots are never stored • Optional browser permissions can be reviewed and revoked from the extension settings • Local data can be exported, deleted, or fully reset at any time Who It's For • Developers building with local and self-hosted AI models • Researchers testing different LLM providers and model capabilities • Students learning local and offline AI workflows • Teams using private LAN-hosted model servers • Privacy-conscious users who want control over providers and stored data Setup 1. Install the extension 2. Start Ollama, LM Studio, llama.cpp, or another supported provider 3. Open the extension settings and connect using a localhost, LAN, or remote endpoint 4. Select a model and start chatting 5. Optional: turn on the experimental browser agent in Settings to try simple tasks in your tab Important Notes • Ollama Client is a frontend client and does not include AI models • Local inference performance depends on your hardware, selected model, and backend configuration • Custom and remote provider support is marked Beta because compatibility can vary between services • The browser agent is experimental; watch its steps, review each approval, and check results yourself • Remote providers may have their own privacy policies, pricing, authentication, and data-retention rules Page Context, Script Injection and Agent Permissions When you ask the model to use the current page, Ollama Client first contacts its content script on that tab. If the content script is unavailable, for example because the tab was opened before the extension was installed or updated, the extension uses the scripting permission to inject its content script into that specific tab. This injection is: • Triggered only by a user-requested page-context action or agent task • Scoped to the tab involved in the request • Limited to reading the page content needed for that request • Never performed continuously, on a schedule, or across every open tab • Not used on protected browser pages where extensions cannot run The browser agent also uses the debugger permission to send trusted clicks and key presses, capture screenshots for vision models, and handle page dialogs; the webNavigation permission to track which frame and page each action belongs to;…