Add vault context injection, frontmatter tool, vision idea

1 files changed, 66 insertions, 8 deletions

diff --git a/README.md b/README.md
index 0f2e19f..6079304 100644
--- a/README.md
+++ b/README.md
@@ -2,10 +2,67 @@
 
 The Obsidian AI Note Management Plugin — powered by [Ollama](https://ollama.com).
 
+Chat with a local AI that has full access to your vault. Search files, read and edit notes, manage frontmatter, create and move files — all through natural conversation with tool-calling support.
+
+## Features
+
+### AI Chat Sidebar
+
+- Streaming responses with real-time token-by-token display (desktop) or full-response delivery (mobile)
+- Markdown rendering with clickable `[[wiki-links]]` in AI responses
+- Multi-turn conversation with full message history
+- Abort generation mid-stream with a Stop button
+- FAB speed dial for quick access to Settings, Tools, and Clear Chat
+
+### Vault Tools
+
+Enable tools to give the AI direct access to your vault. Each tool can be individually toggled on/off. Destructive actions require user approval before executing.
+
+| Tool | Description | Approval |
+|------|-------------|----------|
+| **Search Files** | Find files by name or path | No |
+| **Read File** | Read file content (includes parsed frontmatter as JSON) | No |
+| **Search Contents** | Grep-style text search across all markdown files | No |
+| **Get Current Note** | Get the path of the currently open note | No |
+| **Edit File** | Find-and-replace text in a file | Yes |
+| **Create File** | Create a new note (auto-creates parent folders) | Yes |
+| **Delete File** | Move a file to system trash | Yes |
+| **Move/Rename File** | Move or rename a file (auto-updates all links) | Yes |
+| **Set Frontmatter** | Add, update, or remove YAML frontmatter properties | Yes |
+
+The AI follows a mandatory read-before-edit workflow to prevent data loss. Approval dialogs show full diffs for review before any changes are applied.
+
+### Vault Context Injection
+
+Optionally inject a summary of your vault structure into every conversation so the AI understands your vault without searching first:
+
+- **Folder tree** — indented ASCII tree of all vault folders
+- **Tag taxonomy** — all tags sorted by usage count
+- **Recent files** — most recently modified notes (configurable count)
+- **Stats** — vault name, total notes, total folders
+
+All data comes from the metadata cache — no file reads, instant computation.
+
+### Custom System Prompt
+
+Point the plugin at any vault note to use as persistent AI instructions (writing style, formatting rules, etc.). The note content is injected into the system prompt alongside tool instructions and vault context.
+
+### Settings
+
+- **Ollama URL** — connect to any Ollama instance (localhost, LAN IP, remote)
+- **Model selection** — auto-populated from connected Ollama server
+- **Temperature** — control response randomness (0–2)
+- **Context window** — set `num_ctx` with model max display and one-click apply
+- **Max output tokens** — set `num_predict` (-1 for unlimited)
+
+### Mobile Support
+
+Fully functional on Obsidian Mobile. Uses Obsidian's `requestUrl()` for network requests (non-streaming fallback) to reach Ollama over LAN. Set the Ollama URL to your computer's LAN IP instead of localhost.
+
 ## Prerequisites
 
 - [Obsidian](https://obsidian.md) v0.15.0 or later
-- [Ollama](https://ollama.com) installed and running locally (default: `http://localhost:11434`)
+- [Ollama](https://ollama.com) installed and running (default: `http://localhost:11434`)
 - [Node.js](https://nodejs.org) v16 or later (for building from source)
 
 ## Building from Source
@@ -45,16 +102,17 @@ ln -s /path/to/ai-pulse ai-pulse
 
 Then run `npm run dev` and reload Obsidian to pick up changes.
 
-## Usage
+## Quick Start
 
-1. Click the **message icon** in the left ribbon or run the **"Open AI Chat"** command from the command palette.
-2. The chat sidebar opens in the right panel.
-3. In the **Settings** section at the bottom of the sidebar:
-   - Set the **Ollama URL** (defaults to `http://localhost:11434`).
-   - Click **Test** to verify the connection.
+1. Click the **message icon** in the left ribbon or run **"Open AI Chat"** from the command palette.
+2. Click the **gear icon** (FAB) → **AI Settings**:
+   - Set the **Ollama URL** if not using the default.
+   - Click **Connect** to verify the connection and load models.
    - Select a **Model** from the dropdown.
-4. Type a message and press **Enter** to chat with the AI.
+3. Click **Tools** to enable vault tools (optional but recommended).
+4. Type a message and press **Enter** to chat.
    - **Shift+Enter** inserts a newline.
+   - Click **Stop** to abort a streaming response.
 
 ## License