The Private AI Engineering Platform

Desktop v1.5.26

VS Code v1.5.1

Deploy Agents on Your Code, Knowledge & Business — Privately

Tanvrit AI is the private AI engineering platform and an agent platform: named, verified Engineering Agents, an in-app Marketplace, and Domain Packs that operate real business systems, all on a local-first runtime with deep repository intelligence. Your code never leaves your machine; every agent action is staged, reviewable, and reversible.

🖥 Download for macOS Or open the web app →

The Agent Platform — desktop v1.5.26: four named Engineering Agents (Code Reviewer · Test Writer · Migration · Docs) · in-app Agent Marketplace · Domain Packs with the Commerce pack + Store Operations Agent · headless agent CLI + scheduled runs on Tanvrit Compute · run retention + bearer-hardened local API. v1.5.26 adds Context Pack grounding on every agent surface + self-healing tool-retry; VS Code extension v1.5.1 pairs with it fully. explore the platform → · changelog →

~45

MCP Tools

Languages

AI Providers

90%

Token Savings

How It Works

Three steps to supercharge any AI with your codebase

📂

Index Your Project

Point Tanvrit AI at any directory. It parses 19 languages, extracts symbols, builds dependency graphs, and stores everything locally.

🔌

Connect Your AI Client

Configure Claude Desktop, Cursor, or any MCP-compatible client to use Tanvrit AI's MCP server at localhost:19281.

💬

Ask Anything

Your AI client now has deep code intelligence. Ask it to find dead code, trace call graphs, analyze architecture, or explain any function.

The Platform

Six Pillars, One Local Core

From provider routing to a VS Code copilot — everything runs against a local index that never leaves your machine.

🔀Shipped

AI Router

One routing layer over 7 providers — automatic fallback, circuit breakers, and a cost-aware registry of ~25 models. The client router is shipped and the AI gateway is live for org policy and metering; hosted inference (Tanvrit-managed keys) is rolling out.

🤖Shipped

Agent Runtime

Named, scope-pinned agents on a runtime with staged edits — apply or reject every hunk — self-verification with rollback, run history with token and cost attribution, headless CLI runs, and scheduled agents on Tanvrit Compute.

🗂Shipped

Repository Intelligence

A 19-language indexer with dependency and call graphs and ~45 built-in MCP tools (per-pack tools register when you enable a Domain Pack). Hybrid search — FTS5 full-text fused with real semantic embeddings over your code — plus the semantic_search tool.

👥Shipped — opt-in

Team Knowledge

Opt-in sync of curated decisions, rationales, and annotations across your team — default off, every payload redacted. The raw code index never leaves your machine; the hosted sync backend is live.

🏢Server-side live

Enterprise

Org allowlists, enforced local-only mode, and BYOK keys encrypted at rest ship in the client. Seats, entitlements, and marketplace curation are live on the Tanvrit server; a self-host bundle is available in preview (evaluation mode). Hosted SaaS onboarding is rolling out.

🧩Live

VS Code Copilot

Streaming agent runs in a panel, Pending Edits review from the editor (the extension never writes files itself), context sync, and run history. v1.5.1 pairs fully with desktop v1.5.26.

Explore the platform in depth →

The Agent Platform

Agents That Do Real Work

Named, versioned, documented — each agent runs with a tool scope enforced by the runtime, stages every edit for your review, and attributes every token it spends. Agents act; you decide.

Code Reviewer

v1.0.0

read-only

Diff-first review with blast-radius analysis and severity-ranked file:line findings — plus an explicit “Not checked” list. Produces findings, never edits.

Test Writer

v1.0.0

staged writes · verified

Finds untested changed code, mirrors your project's own test style, and runs the suite to green — a headless run fails (and rolls back) without passing verification.

Migration Agent

v1.0.0

staged writes

Exhaustive from→to migrations: triple-search site discovery, staged per-site edits, and a changed/skipped summary table that must reconcile with discovery.

Docs Agent

v1.0.0

Markdown-jailed

Keeps documentation truthful against the diff — never documents a claim it can't trace to code. Reads anything, writes only Markdown.

Store Operations Agent

v1.0.0

commerce-scoped

Works the order queue — investigates with catalog, customer, and analytics lookups, then proposes confirm / decline / inventory actions. Every money-affecting call needs your approval.

Agent Marketplace

in-app

curated

Install agents and Domain Packs in one click from the in-app Marketplace — curated, reviewed, scope-pinned. Installs are validated and provenance-stamped; a download can never shadow first-party code.

Domain Packs bind business systems to the same runtime — the Commerce pack v1 ships 12 governed commerce_* tools, opt-in and approval-gated. How agents are governed →

Model Context Protocol

~45 Precision Tools

Every tool is designed to give AI models the exact context they need — nothing more, nothing less.

search_code

Full-text code search with context

get_chunk90% tokens

Get function/class body only

get_file_structure95% tokens

Symbol outline, no bodies

get_call_graph

Function callers and callees

analyze_inheritance

Type hierarchy traversal

find_references

All usages of a symbol

analyze_impact

Blast radius of a change

find_dead_code

Unreferenced symbols

get_metrics

Complexity and quality metrics

git_blame

Author per line with commit

analyze_dependencies

Import graph analysis

search_symbols

Find classes, functions, interfaces

+ ~30 more tools including get_context, get_imports, search_definitions, get_project_overview, get_file, and more — ~45 built-in, plus per-pack tools (e.g. commerce_*) that register when you enable a Domain Pack.

One Codebase, Four Platforms

Built with Kotlin Multiplatform and Compose Multiplatform. Native performance on every device.

🖥

macOS

Download

Native desktop app with full MCP server on port 19281. Requires macOS 13 Ventura or later.

.dmg · 200 MB

🪟

Windows

Download

Native Windows installer with full MCP server on port 19281. Requires Windows 10 or later.

.msi · ~225 MB

🐧

Linux

Download

Debian package with full MCP server on port 19281. Requires Ubuntu 20.04+ or compatible.

.deb · ~215 MB

🌐

Web

Launch

Browser-based app via Kotlin/WASM for instant access — no install required.

No install

10 Built-in Dev Prompts

Ready-to-use prompts for common developer workflows, powered by your indexed codebase.

code-review

explain-code

refactor

write-tests

document

find-bugs

optimize

architecture-review

security-audit

migration-guide

Architecture

Built for Developers, by Developers

Tanvrit AI stores everything locally — your code never leaves your machine. The MCP server runs at localhost:19281, giving any AI client deep access to your codebase while keeping you in control.

100% local — no cloud uploads
SQLite database, fully inspectable
Incremental indexing, hash-based
11 database tables for deep analysis

MCP ServerKtor CIO · port 19281

DatabaseSQLDelight · SQLite · 11 tables

Parsers19 languages · ANTLR-based

AI ProvidersClaude · GPT · Gemini · Ollama

PlatformsAndroid · iOS · Desktop · Web

Built for Real Work

What Developers Do With Tanvrit

From onboarding to refactoring, Tanvrit AI gives your AI the context it needs to actually help — instead of hallucinating.

Onboarding

Ramp up on a new codebase in hours, not weeks

Ask your AI “how does auth work here?” and it traces the real call graph instead of guessing from filenames. New hires ship their first PR in day one.

Refactoring

Safe refactors with real blast-radius analysis

Before renaming a function, see every caller, every test, and every cross-module reference. No more “oops, that was used in 14 places.”

Cost control

Stop paying Claude to re-read your repo

MCP tools return only the exact chunk or symbol the model needs. Teams report 90% token savings vs. dumping files into context windows.

Dead code

Find and delete unreferenced symbols in minutes

Unreferenced functions, orphan classes, unused types — Tanvrit builds the real reference graph so you can cut cruft with confidence.

Code review

Catch architectural drift before it ships

Feed PR diffs plus dependency/metric context into your AI reviewer. Get feedback grounded in your actual architecture, not generic style rules.

Agent tools

Build agents that actually understand your code

Any MCP-compatible agent — Claude Desktop, Cursor, a custom agent — plugs into ~45 tools on port 19281. Bring your own LLM.

Why Tanvrit

You have options. Here's how we compare.

Capability	Tanvrit AI	Cloud indexers	Plain LLM
100% local, no cloud uploads	Yes	No	—
Works with any AI client via MCP	Yes	Limited	No
Real call graph & dependency analysis	Yes	Some	No
19 languages parsed	Yes	Varies	—
Token savings on AI queries	90–95%	Varies	0%
One-time price, self-hosted option	Yes	Rarely	—

Simple Pricing

Start Free. Scale When Ready.

Your code stays on your machine. Pay only when you need team features or hosted compute.

Free

Local-only, no quotas

$0/ forever

✓All ~45 MCP tools
✓All 19 language parsers
✓Local Ollama models
✓Unlimited local repos
✓Community support

Download

Plus

Bring your own API key

$9/ month

✓Everything in Free
✓Bring-your-own Anthropic / OpenAI / Gemini key
✓Cloud fallback for huge codebases
✓Provider routing & cost dashboard
✓Priority email support

Get Started

Team

Shared MCP server for the team

$29/ user / month

✓Everything in Plus
✓Shared MCP server
✓Audit logs
✓SSO
✓Org-wide prompt & tool policies
✓Onboarding

Talk to us

The free tier is the trial — your source code stays on your machine. Need air-gapped or self-hosted? See full pricing →

Frequently Asked

Does my code leave my machine?+

No. Indexing, parsing, and MCP tool execution happen 100% locally. The only network calls are to whichever AI provider (Claude, OpenAI, Gemini, etc.) you configure — and those only receive the specific chunks your agent requests.

Which AI clients work with Tanvrit?+

Any MCP-compatible client: Claude Desktop, Cursor, Zed, or a custom agent you build. Tanvrit exposes an MCP server on localhost:19281 with ~45 tools. We also bundle an AI chat UI with Claude/OpenAI/Gemini/Ollama support.

How big a codebase can it handle?+

Tested on repos up to 1M lines of code. SQLite scales comfortably into the millions of symbols. For very large monorepos, use project-level directories — index only what you're working on.

Do I need to bring my own API keys?+

Not for local use — the Free tier runs entirely on local Ollama models, no key required. Add your own Anthropic, OpenAI, Gemini, Groq, Mistral, or Ollama key in settings when you want cloud-quality generation; Plus adds automatic cloud fallback for codebases too large for the local model.

Can I self-host the Team plan?+

Yes — Team customers can deploy Tanvrit's sync layer on their own infrastructure. Source code and queries stay entirely inside your network.

What languages are supported?+

19 and growing: Kotlin, Swift, TypeScript, JavaScript, Python, Go, Rust, Java, C, C++, C#, Dart, Scala, Ruby, PHP, Elixir, Haskell, Lua, Shell. Ask for yours.

Is there a free trial?+

The free tier is the trial — the full product runs locally, forever, at $0. No credit card to start. Upgrade to Plus only when you actually want cloud fallback; we don't believe in fake limits to push upgrades.

How does the token savings work?+

Instead of your agent dumping an entire file into the context window, Tanvrit's tools return only the symbol, chunk, or structure being asked about. Typical savings: 90–95% per AI call on large files.

Put Agents to Work on Your Code

Start indexing in minutes. No signup. No cloud. Just your code and your agents.

Launch Web App 🖥 macOS 🪟 Windows 🐧 Linux