Vidurai: Local-First AI Memory Layer

Vidurai (v2.2.0) is a local background service that indexes your development context. It bridges your local codebase with AI tools (like Claude, Cursor, and VS Code) without uploading your source code to third-party vector clouds.

It runs locally, stores data locally (~/.vidurai), and serves context via the Model Context Protocol (MCP) and IPC.

विस्मृति भी विद्या है (Forgetting too is knowledge) — The system intelligently compresses and forgets low-value context while preserving critical insights.

Installation

Step 1: Install the Brain

We recommend pipx to isolate dependencies.

pipx install vidurai
# OR
pip install vidurai

Step 2: Start the Guardian

vidurai start
# Output: Vidurai Guardian started (PID: 12345)

Verify Installation

vidurai --version
vidurai status

Integrations

1. Visual Studio Code (Primary)

The extension syncs active file context and terminal errors.

• Install: Search for "Vidurai" in the VS Code Marketplace.
• Handshake: Automatically connects to the daemon on startup.

2. Cursor Editor (Native Support)

Cursor works natively via the VS Code extension ecosystem.

1. Download vidurai-extension.vsix from Releases.
2. Install via Terminal (Recommended for reliability):

   cursor --install-extension vidurai-extension.vsix
   

3. Claude Desktop (MCP)

Vidurai implements the Model Context Protocol.

Config Location:

• Linux/Mac: ~/.config/claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Config:

{
  "mcpServers": {
    "vidurai": {
      "command": "vidurai",
      "args": ["server"]
    }
  }
}

Or use the auto-installer:

vidurai mcp-install

CLI Reference (v2.2.0)

Full reference: docs/CLI.md (auto-generated via AST static analysis)

Daemon Control

Memory Operations

Strategic Forgetting (SF-V2)

Import & Export

Advanced & Debug

Architecture

+-------------------------------------------------------------+
|                      YOUR MACHINE                           |
+-------------------------------------------------------------+
|                                                             |
|   +-----------+         +-----------------------------+     |
|   |  VS Code  |<--IPC-->|                             |     |
|   | Extension |         |    Vidurai Guardian         |     |
|   +-----------+         |    (Background Daemon)      |     |
|                         |                             |     |
|   +-----------+         |  +-----------------------+  |     |
|   |  Claude   |<--MCP-->|  |  Context Mediator    |  |     |
|   |  Desktop  |         |  |  (Semantic Compress) |  |     |
|   +-----------+         |  +-----------------------+  |     |
|                         |            |                |     |
|   +-----------+         |  +---------v-----------+    |     |
|   |  Cursor   |<--IPC-->|  |  Memory Store       |    |     |
|   |  Editor   |         |  |  (SQLite + Parquet) |    |     |
|   +-----------+         |  +---------------------+    |     |
|                         |                             |     |
|                         +-----------------------------+     |
|                                                             |
+-------------------------------------------------------------+

Storage Stack

• Hot Storage: SQLite (Metadata) + sqlite-vec (Embeddings)
• Cold Storage: Apache Parquet (Archives)
• Privacy: All data in $HOME/.vidurai. PII (API Keys) redacted via Regex Gatekeeper.
• Learning: Offline Reinforcement Learning (RL) prunes low-salience context.

Data Directory

~/.vidurai/
├── memory.db          # SQLite database (hot storage)
├── vidurai.log        # Daemon logs (rotated, max 50MB)
├── daemon.pid         # Process ID file
└── archive/           # Parquet files (cold storage)
    └── YYYY/MM/       # Partitioned by date

Quick Verification

# Check daemon status
vidurai status

# View recent logs
vidurai logs -n 20

# Search memories
vidurai recall --query "authentication"

# Check memory health
vidurai hygiene

Documentation

License

MIT License. See LICENSE for details.

Vidurai — Because your AI should remember what you've already solved.