Summarizing a search

Run a search using SearXNG as usual, then click the ✨ toolbar button. A side panel opens and streams a summary of the top results, with clickable citations that open the source in a new tab.

Ask anything

Press Ctrl+Shift+A (or ☰ → Ask local AI…) to open the same panel as a general chat. It's multi-turn, so follow-up questions keep their context; New chat clears it and starts over. You can even follow up on a summary with a question — the summary's content carries into the conversation.

A model dropdown at the top of the panel lists whatever models are installed in your local Ollama, refreshed every time the panel opens.

Search web

A Search web checkbox beside the chat question field is off by default. When it's on, your latest typed question — and only that question, never the rest of the conversation, your browsing history, or any open page — is sent through your local SearXNG instance to its upstream search engines. The model then receives up to six result snippets alongside your question, so it can answer with current information and cite clickable sources, instead of relying only on what it already knows. The model doesn't browse pages on its own; it only sees the snippets it's handed. A search failure is reported explicitly rather than silently falling back to no results. Stop cancels both an in-progress search and the model's reply. Your choice is remembered for next time; unchecking it returns chat to normal, no-web-access behavior.

What's sent, and to where

For summaries, Vodou reads the top results from the rendered SearXNG page and sends them, with your query, to Ollama running on your own machine (127.0.0.1). In chat mode, only what you type is sent — never the page you're on, its address, or your history — with one deliberate, explicit exception: see Checking whether a site is safe. With Search web enabled, your latest question additionally goes out through SearXNG to its configured search engines — the only outbound step in this feature that leaves your local Ollama instance. Configuration (which model, the endpoint, how many past messages to resend, the Search web toggle itself) lives in ~/.vodou/ai_search.json.