Tanvrit AI

Local AI code intelligence for VS Code. Index your codebase, get inline completions and an AI chat panel, and connect any AI model — Claude, GPT-4, Gemini, or Ollama — to your code via MCP. Everything runs on your machine.

What it does

This extension connects VS Code to the Tanvrit AI desktop app running on your machine. The desktop app indexes your codebase locally and exposes two services the extension auto-detects:

An OpenAI-compatible local API on localhost:19280 — powers inline completions, the chat panel, and the Explain / Fix / Generate / Refactor commands
An MCP server on localhost:19281 — 31 code-intelligence tools any MCP client (Claude Desktop, Cursor, Zed) can call

No cloud indexing. No code leaving your machine.

Features

Inline completions — Copilot-style ghost-text suggestions powered by local models
AI chat panel — streaming chat about your code in the activity bar (Cmd+Shift+I)
Explain / Fix / Generate / Refactor — one keystroke or right-click on any selection
Run Agent on Selection — describe a change in plain English, let the agent do it
31 MCP tools — call graphs, dependency graphs, type hierarchies, dead-code detection, git blame, metrics, and more
19 language parsers — Kotlin, Swift, TypeScript, Python, Go, Rust, Java, C, C#, Dart, Scala, Ruby, PHP, and more
One-click Claude Desktop setup — the extension writes your MCP config (with bearer token) automatically, backing up any existing file
Copy MCP Config — grab a ready-to-paste snippet for Cursor, Zed, or any MCP client
Bring your own model — Claude, GPT-4, Gemini, DeepSeek, Groq, Mistral, or fully offline with Ollama / the bundled local engine

Requirements

Install the Tanvrit AI desktop app — it runs the local engine and MCP server this extension connects to.

Download: ai.tanvrit.com/download (macOS, Linux)

If the app isn't running, the extension shows a one-time prompt with the download link; all features light up automatically once the app is detected.

Quick start

Download and launch Tanvrit AI from ai.tanvrit.com/download
Open any project in VS Code — the extension detects the running app
Press Cmd+Shift+I / Ctrl+Shift+I to open the chat panel
Optional: accept the one-time prompt to auto-configure Claude Desktop with Tanvrit's MCP tools

Keyboard shortcuts

Action	Mac	Windows / Linux
Open Chat	`Cmd+Shift+I`	`Ctrl+Shift+I`
Explain Selection	`Cmd+Shift+E`	`Ctrl+Shift+E`
Fix Selection	`Cmd+Shift+X`	`Ctrl+Shift+X`
Generate Code	`Cmd+Shift+G`	`Ctrl+Shift+G`
Toggle Inline Completions	`Cmd+Shift+A`	`Ctrl+Shift+A`
Add to Context	`Cmd+Shift+K`	`Ctrl+Shift+K`
Run Agent on Selection	`Cmd+Shift+R`	`Ctrl+Shift+R`
Switch Model	`Cmd+Shift+M`	`Ctrl+Shift+M`

Right-click any selection for Explain, Fix, Generate, Add to Context, and Run Agent in the context menu.

Commands

All commands are available via the Command Palette (Cmd+Shift+P):

Command	Description
`Tanvrit AI: Explain Selection`	Explain what the selected code does
`Tanvrit AI: Fix Selection`	Identify and fix bugs in selected code
`Tanvrit AI: Generate Code`	Describe what you need, get code at the cursor
`Tanvrit AI: Refactor Selection`	Extract functions, simplify logic, improve naming
`Tanvrit AI: Open Chat Panel`	Open the chat panel
`Tanvrit AI: Add to Context`	Add the current file or selection to the chat context
`Tanvrit AI: Run Agent on Selection`	Describe a change; the agent applies it
`Tanvrit AI: Switch Model`	Pick the active chat model
`Tanvrit AI: Load Model`	Load a model into the local engine
`Tanvrit AI: Show Engine Status`	View engine health, loaded models, memory
`Tanvrit AI: Copy MCP Config`	Copy an MCP config snippet (stdio or SSE) for any client
`Tanvrit AI: Set Up Claude Desktop (auto-config)`	Write Claude Desktop's MCP config automatically
`Tanvrit AI: Toggle Inline Completions`	Enable or disable inline suggestions

Settings

Setting	Default	Description
`tanvrit.server.port`	`19280`	Port for the local Tanvrit AI engine
`tanvrit.completion.enabled`	`true`	Enable inline completions
`tanvrit.completion.debounceMs`	`150`	Debounce delay before requesting a completion
`tanvrit.completion.maxTokens`	`128`	Max tokens per completion
`tanvrit.completion.temperature`	`0.2`	Sampling temperature
`tanvrit.chat.maxTokens`	`2048`	Max tokens for chat responses
`tanvrit.model.autocomplete`	`qwen25-coder-1.5b`	Model for inline completions
`tanvrit.model.chat`	`deepseek-coder-v2-lite`	Model for chat

Privacy

Everything runs locally. The extension only talks to localhost (ports 19280/19281) — no cloud API calls, no telemetry, no data collection. Your code never leaves your machine.

Support

Tanvrit Private Limited · tanvrit.com