---

Smart Notes

An Obsidian plugin that turns your vault into an AI-assisted knowledge workspace. Chat with your notes, optimize them, autocomplete inside the editor, and give the AI a structured view of your vault — all powered by the model provider you configure.

Display name: Smart Notes · Plugin ID: smart-notes · Requires Obsidian 1.11.4+

Features

AI chat panel

A dedicated right-sidebar chat about your vault and the active note.

• Multiple sessions — create, switch, rename, and delete conversations; collapse/expand the history sidebar. Each session keeps its own messages, attachments, skills, and web-search state.
• Streaming replies with live Markdown rendering, a typing indicator, and per-reply token usage.
• Per-conversation model & role — pick the model (grouped by link) and the role for each chat; the choice stays scoped to that conversation.
• Reasoning display — for reasoning models (e.g. DeepSeek-R1, OpenAI o1) the model's thinking process is shown in a collapsible block above the reply (toggle in settings).
• History window — older turns are compressed into a summary so long chats don't blow the token budget.

Model configuration (multi-provider)

Configure one or more model links, each pointing at an AI backend:

• OpenAI-compatible (OpenAI, Azure OpenAI, DeepSeek, local gateways, …) — Base URL + API key.
• Ollama (local) — Base URL only, fully offline if the model runs locally.
• Each link can list multiple models and carry optional per-link max tokens / temperature.
• API keys are stored in Obsidian's secret storage (keychain), not in plain plugin data. A Test connection button verifies a link before saving.
• Set any link as the default; the active link is used by every AI feature.

Roles

Define reusable system-prompt roles (name + prompt, with an optional avatar). A default role is injected into the system prompt of every AI feature (chat, optimize, autoprompt, frontmatter). In a chat you can switch the active role per conversation. This replaces the old single "custom instructions" field.

Knowledge base & privacy model

The plugin is designed so the AI only sees what you explicitly share:

• File-path index (optional) — injects a list of all Markdown file paths so the AI knows what's in your vault, without loading any content.
• Frontmatter index (optional) — injects each note's YAML properties (tags, category, summary, …) so the AI understands your vault's structure, again without body text.
• Attachments — attach specific files or whole folders to a conversation; only their content is read.
• @-mentions — if you name a vault file in a message, that file's content is pulled in.
• Everything else stays local. You can hide the "Attach file/folder" button entirely.

Skills

Drop skill bundles (a folder containing SKILL.md) into the AI folder's skills/ directory. Skills can be toggled on for new chats, and managed per conversation via the chat's Skill button. Only active skills' content is sent to the AI; inactive skills are listed by name so the AI knows they exist and can suggest enabling them.

Web search

Optional web search inside a chat (🌐 button, per session). Supports Tavily / Serper (Google) / Brave / SearXNG, with multi-key polling (one key fails → try the next). Results can be cited inline with [n]. Off by default and only runs when you explicitly enable it in a chat.

Note optimization

The Optimize current note command asks the AI to polish wording, restructure, or summarize the active note. Preview the result in a diff-style modal (Original vs Optimized) and apply with one click. Internal-link format ([[wikilink]] vs [markdown](path.md)) and path style are configurable.

Frontmatter generation

The Generate frontmatter command reads the current note and asks the AI to produce a YAML frontmatter block (title, date, tags, category, summary, keywords, …). A custom template can override the default system prompt.

In-editor autocompletion

• Realtime autoprompt (optional, off by default) — debounced inline "ghost text" suggestions as you type; accept with Tab.
• Autoprompt at cursor command — trigger a suggestion on demand.

Conversation memory & long-term profile

• All chat sessions are persisted in your vault, so conversations survive restarts.
• In the background the plugin distills your chat history into a long-term profile memory (memory/MEMORY.md + a daily note) along configurable dimensions (occupation, tech stack, preferences, …). It is injected as background context and is fully user-editable.

Localization

The UI is fully internationalized and follows your Obsidian language (Settings → About → Language). English and 简体中文 are bundled. To add a language, copy src/i18n/locales/en.json, translate the values, and register it in src/i18n/index.ts.

Privacy & network usage

This plugin makes network requests to the endpoints you configure:

• Your chosen model link (OpenAI-compatible or Ollama) — note text is sent to the chat/completions endpoint you set.
• Optional web search providers (Tavily / Serper / Brave / SearXNG) — only when you enable web search in a chat.

No data is sent anywhere except the endpoints you explicitly configure. There is no telemetry. API keys live in Obsidian's secret storage, not in plain plugin data. Vault file contents are only transmitted when you attach a file/folder or mention it by name in a message; indexes carry paths/metadata only.

Settings

Section	Setting	Description
Model configuration	Model links	Add / edit / delete / set default AI backends; each holds type, Base URL, API key (keychain), models, optional max tokens & temperature.
	Test connection	Verify a link with a tiny ping.
	Data storage path	Name of the vault-root folder for memory & skills (default .smartnotes).
Interaction	Realtime autoprompt	Inline ghost-text suggestions while typing (off by default).
	Realtime debounce (ms)	Idle time before a realtime suggestion is requested.
	Optimize current note	Enable the optimize command.
	Open AI chat panel	Enable the chat command.
	Attach current note to chat	Auto-attach the active note when opening chat (off by default).
	Show AI reasoning	Show reasoning models' thinking block.
	Generate frontmatter	Enable the frontmatter command.
	Frontmatter template	Custom system prompt for frontmatter generation.
	Internal link format / type	Wikilink vs Markdown link; shortest / relative / absolute paths.
	Max tokens / Temperature	Global generation limits (per-link values override).
Knowledge base	Enable file selection	Show the "Attach file/folder" button in chat.
	Enable file index	Inject all Markdown file paths into the system prompt (off by default).
	Max files in file index	Cap injected paths (0 = unlimited).
	Enable property index	Inject Frontmatter metadata into the system prompt (off by default).
	Properties to index / Max files / Max chars	Fine-tune the Frontmatter index.
	Max chars per attached file	Truncate attached content sent to the AI.
	History window	Recent N messages sent to the model; older turns compressed to a summary (0 = send all).
User profile	Enable user profile memory	Build & inject long-term profile memory.
	Profile dimensions	One dimension per line used to extract the profile.
	Memory file (MEMORY.md)	View/edit the AI-curated profile; auto-saved.
	Profile memory char cap	Truncate profile before injection.
Roles	Enable roles	Show the Role button and inject the default role.
	Role info	Add / edit / delete roles; set a default.
Skills	Enable skills	Show the Skill button and allow injection.
	Default skills	Toggle which skills are enabled for new chats.
Web search	Enable web search	Master switch for web search.
	Search provider	Tavily / Serper / Brave / SearXNG.
	API keys / instances	Per-provider credentials (multi-key polling).
	Max results / Max chars / Show citations	Search tuning.

AI folder layout

On load the plugin creates the following at your vault root (folder name configurable, default .smartnotes):

.smartnotes/
├── sessions/
│   ├── index.json            # session index (metadata only: id, title, time, message count)
│   └── session-<id>.json     # one conversation's full content (messages, attachments, skills, web search)
├── memory/
│   ├── MEMORY.md             # long-term profile memory (auto-built, editable)
│   └── yyyy-mm-dd.md         # daily memory note
└── skills/
    └── README.md            # what this folder is for

• The index holds only lightweight metadata; the last 10 sessions are fully loaded on startup, older ones are read lazily when you open them.
• Each per-session file is written only when that session changes.

Commands

• Optimize current note
• Autoprompt at cursor
• Open AI chat panel
• Generate frontmatter

Install (development)

1. Copy the built main.js, manifest.json, and styles.css into <your-test-vault>/.obsidian/plugins/smart-notes/.
2. Enable the plugin in Obsidian settings.
3. Configure a model link (Settings → Smart Notes → Model configuration) and add your API key / Ollama URL.

Use a separate test vault while developing — do not run this on your main vault.

License

MIT © hellokunzai