Obsidian Stats logoObsidian Stats

AI LLM Search

by darklii
5
4
3
2
1
Score: 50/100
Install Code

Description

Use your own model API to search and chat with selections, notes, your Obsidian knowledge base, and the web.

Reviews

No reviews yet.

Stats

0
stars
20
downloads
0
forks
6
days
6
days
6
days
0
total PRs
0
open PRs
0
closed PRs
0
merged PRs
0
total issues
0
open issues
0
closed issues
11
commits

Latest Version

1.1.5
6 days ago

Changelog

What changed

  • Search selected topics instead of generic pronouns such as “这是什么”.
  • Filter unrelated web candidates, including exact acronym matching for cases such as NAIC versus NAICS.
  • Group direct context, vault retrieval, and web candidates so users can see exactly what was sent to the model.
  • Improve source titles, paths, hostnames, and narrow-sidebar readability.
  • Add regression tests for context-query construction and source relevance.
See all version on GitHub

README file from

Github

AI LLM Search

AI LLM Search is a Chinese-first AI sidebar for Obsidian. It lets you use your own OpenAI-compatible model API to chat with selected text, the current note, relevant notes from your knowledge base, and optional web search results without leaving your reading workflow.

Features

  • Ask questions about the current note.
  • Select text in editing or reading view, then right-click to ask AI or search the whole vault.
  • Select text in Obsidian's built-in Web Viewer, then right-click to ask AI with the page URL as source context (desktop).
  • Search Markdown notes locally with BM25; no embedding service is required.
  • Combine current-note, whole-vault, and web context in one conversation.
  • Show exactly which selections, current notes, vault results, and web candidates were sent to the model, grouped by context type.
  • Use DuckDuckGo without a search API key, or configure Tavily.
  • Connect to OpenAI-compatible services, including local endpoints.
  • Store API keys in Obsidian SecretStorage rather than the plugin data file.
  • Adapt the sidebar to narrow desktop panes and mobile layouts.

The interface is currently available in Simplified Chinese.

Installation

Community plugins

  1. Open Settings → Community plugins → Browse.
  2. Search for AI LLM Search.
  3. Install and enable it.

Manual installation

  1. Download main.js, manifest.json, and styles.css from the latest GitHub release.
  2. Create <knowledge-base>/.obsidian/plugins/ai-llm-search/.
  3. Copy the three files into that directory.
  4. Reload Obsidian and enable AI LLM Search under Community plugins.

Configuration

  1. Open Settings → AI LLM Search.
  2. Enter an OpenAI-compatible API base URL and model ID.
  3. Select or create the model API key with Obsidian SecretStorage. A key is optional for local endpoints such as localhost.
  4. Optionally limit the folders included in local vault search.
  5. Optionally choose DuckDuckGo or Tavily for web search.
  6. Use the ribbon icon to open the AI sidebar.

The command 针对划线内容提问 is available in the command palette. Obsidian community guidelines do not allow plugins to assign default hotkeys, so you can bind your preferred shortcut under Settings → Hotkeys.

Web Viewer questions

  • In Obsidian's built-in Web Viewer on desktop, select text on the page, right-click, then choose an AI action. The selected text and page URL appear as pending context in the sidebar.

The built-in Web Viewer selection hook uses Obsidian desktop's embedded webview and is therefore desktop-only. The rest of the plugin remains mobile-compatible.

How context is handled

  • Current note: sends the active Markdown note when that scope is enabled.
  • Whole vault: builds a local in-memory BM25 index and sends only the highest-ranked excerpts for the selected text plus the current question.
  • Selected text: sends the highlighted text as additional context.
  • Web: searches for the selected topic plus the current question, filters unrelated candidates, and may fetch URLs included in the question.

Vault indexing happens locally. The plugin does not upload the entire vault unless the selected context itself contains that content.

Privacy and network disclosure

This plugin has no telemetry, analytics, advertising, or developer-operated server.

When you send a message, the question and enabled note excerpts are transmitted directly to the model API endpoint you configured. If web search is enabled, the query is sent to DuckDuckGo or Tavily, and relevant public pages may be fetched directly. API providers and websites process data under their own privacy policies.

Model and Tavily keys are read from Obsidian SecretStorage. They are not stored in the plugin repository or data.json. Non-secret preferences such as the API base URL, model ID, folder filters, and selected SecretStorage entry names are stored through Obsidian's plugin settings API.

Review sensitive note content before sending it to any third-party model or search service.

Development

npm install
npm run build

The production build generates the release asset main.js.

Support

Please report bugs or request features through GitHub Issues.

License

MIT

Sponsors