TT-VSCode-Toolkit

June 23, 2026 · View on GitHub

Interactive learning and development tools for Tenstorrent AI accelerators

VS Code Marketplace License VSCode

Learn Tenstorrent hardware and software through 48 interactive lessons with guided hands-on exercises, production-ready code templates, and intelligent hardware detection. Perfect for developers new to Tenstorrent hardware and teams building production AI inference and custom training pipelines.

Screen capture of TT-VSCode-Toolkit in action

Overview

The TT-VSCode-Toolkit is an educational extension that provides:

  • ✅ 48 Interactive Lessons - From hardware detection to custom training, organized by skill level
  • ✅ Custom Training Ready - Train models from scratch or fine-tune existing ones (validated on hardware!)
  • ✅ Click-to-Run Commands - Execute lessons step-by-step without copy-pasting commands
  • ✅ Hardware Auto-Detection - Automatically detects your Tenstorrent device and provides tailored guidance
  • ✅ Production Templates - Real, tested code you can customize for your projects
  • ✅ Multi-Framework Support - Learn vLLM, TT-Forge™, TT-XLA, and TT-Metalium™
  • ✅ Live Device Monitoring - Real-time temperature, power, and health status in the status bar

Target Audience:

  • Developers new to Tenstorrent hardware
  • AI engineers deploying models on TT accelerators
  • Teams building production inference pipelines
  • ML researchers training custom models
  • Contributors to the Tenstorrent ecosystem

Quick Start

Try in Docker (No Installation)

Run the IDE locally in your browser:

docker run -d -p 8080:8080 -e PASSWORD=demo \
  ghcr.io/tenstorrent/tt-vscode-toolkit:latest

Access at: http://localhost:8080 (password: demo)

Deploy to Cloud with Real Hardware

→ See the Deploy to Koyeb lesson for step-by-step instructions deploying to Tenstorrent n300 hardware on Koyeb.

Installation

Prerequisites

Hardware:

  • Tenstorrent accelerator (n150, n300, T3000, p100, p150, or Galaxy)
  • 32GB+ RAM recommended (16GB minimum)
  • 100GB+ free disk space for models

Software:

  • Linux (Ubuntu 20.04+, RHEL 8+, or compatible)
  • Python 3.10+ (3.11 for TT-XLA)
  • VSCode 1.93+
  • TT-Metalium installed and configured

Verify your environment:

tt-smi                                           # Hardware detected?
python3 --version                                # Python 3.10+?
python3 -c "import ttnn; print('✓ Ready')"       # TT-Metalium working?

Installation

code --install-extension Tenstorrent.tt-vscode-toolkit

Or search "TT-VSCode-Toolkit" in the VSCode Extensions panel (Ctrl+Shift+X).

→ Open in VS Code Marketplace

Other install options:

Option 2: Install from VSIX Package

# Download the latest release from GitHub
gh release download --repo tenstorrent/tt-vscode-toolkit --pattern '*.vsix'

# Install extension
code --install-extension tt-vscode-toolkit-*.vsix

Option 3: Build from Source

# Clone repository
git clone https://github.com/tenstorrent/tt-vscode-toolkit.git
cd tt-vscode-toolkit

# Install dependencies
npm install

# Build and package extension
npm run build
npm run package

# Install
code --install-extension tt-vscode-toolkit-*.vsix

First Steps

  1. Open VSCode - The extension activates automatically on startup
  2. Open Tenstorrent Sidebar - Click the Tenstorrent icon in the activity bar
  3. Start Learning - Begin with "Hardware Detection" lesson or open the Welcome page

Configuration: By default, all lessons are visible. To show only validated lessons, disable "Show Unvalidated Lessons" in settings.

Learning Paths

🎯 Beginner Path (4-6 hours)

Perfect for first-time users

1. Hardware Detection      (5 min)  → Verify your hardware setup
2. Verify Installation     (5 min)  → Test TT-Metalium installation
3. Download Model          (30 min) → Get Llama-3.1-8B or Qwen3-0.6B
4. vLLM Production         (20 min) → Launch production server

What you'll learn: Hardware detection, environment verification, model downloading, and production inference serving with OpenAI-compatible API.

🚀 Intermediate Path (6-8 hours)

For experienced developers

1. Hardware Detection      → Verify setup
2. vLLM Production        → Production serving
3. Image Generation       → Stable Diffusion on TT hardware
4. TT-Forge               → PyTorch model compilation
5. Coding Assistant       → Build an AI coding tool

What you'll learn: Production deployment patterns, multi-modal inference (text + images), compiler workflows, and practical AI applications.

💡 Advanced Path (10-15 hours)

For contributors and power users

1. TT-XLA                 → JAX production compiler
2. RISC-V Programming     → Low-level Tensix core programming
3. Bounty Program         → Model bring-up opportunities
4. TT-Metalium Cookbook   → Custom hardware projects

What you'll learn: Advanced compiler usage, low-level hardware programming, model bring-up workflows, and custom kernel development.

Lesson Catalog

39 lessons organized by category. Hardware badges show validated platforms (✅ = tested and working).

👋 Your journey begins here

0 lessons, 0 validated

🚀 Your First Inference

7 lessons, 7 validated

  • Modern Setup with TT-Installer 2.0 — N150 P300C
  • Hardware Detection — N150 P300C
  • Verify Your Setup — N150 P300C
  • Download Model and Run Inference — N150 P300C
  • Interactive Chat with Direct API — N150
  • HTTP API Server with Direct API — N150
  • Build TT-Metalium from Source — N150 P300C

🏭 Serving Models

4 lessons, 3 validated

  • Production Inference with TT-Inference-Server — N150 P100
  • Production Inference with vLLM — N150 P300C
  • Image Generation with Stable Diffusion XL — N150
  • Video Generation via Frame-by-Frame Diffusion — draft

🔧 Compilers & Tools

3 lessons, 0 validated

  • Image Classification with TT-Forge — P300C
  • JAX and PyTorch/XLA on Tenstorrent — P300C
  • Introduction to TT-Lang — draft

🎯 Applications

5 lessons, 4 validated

  • Coding Assistant with Aider — draft
  • Native Video Animation with AnimateDiff — P300C
  • OpenClaw AI Assistant on TT-QuietBox 2 — P300X2
  • Generating Video on TT-QuietBox 2 — P300X2
  • Local AI Agents on TT-QuietBox 2 — P300X2

🎓 Advanced Topics

4 lessons, 1 validated

  • Bounty Program: Model Bring-Up — draft
  • Exploring TT-Metalium — N150 P300C
  • Twenty-and-Ten Things You Can Do with ttsim — draft
  • ttsim QEMU Bridge: Full-System Simulation — draft

🎓 Custom Training

8 lessons, 0 validated

  • Understanding Custom Training — N150
  • Dataset Fundamentals — N150
  • Configuration Patterns — N150
  • Fine-tuning Basics — N150
  • Multi-Device Training — N150
  • Experiment Tracking — N150
  • Model Architecture Basics — N150
  • Training from Scratch — N150

☁️ Deployment

2 lessons, 2 validated

  • Deploy TT-VSCode-Toolkit to Koyeb — N150
  • Deploy Your Work to Koyeb — N150

👨‍🍳 Tenstorrent Cookbook

6 lessons, 6 validated

  • Tenstorrent Cookbook Overview — N150 P300C
  • Recipe 1: Conway's Game of Life — N150 P300C
  • Recipe 2: Audio Signal Processing — N150 P300C
  • Recipe 3: Mandelbrot Fractal Explorer — N150 P300C
  • Recipe 4: Custom Image Filters — N150 P300C
  • Recipe 5: Particle Life Simulator — N150 P300C

🧠 CS Fundamentals

7 lessons, 0 validated

  • Module 1: RISC-V & Computer Architecture — draft
  • Module 2: The Memory Hierarchy — draft
  • Module 3: Parallel Computing — draft
  • Module 4: Networks and Communication — draft
  • Module 5: Synchronization — draft
  • Module 6: Abstraction Layers — draft
  • Module 7: Computational Complexity in Practice — draft

Key Features

Intelligent Hardware Detection

  • Auto-detects device type (n150, n300, T3000, p100, p150, Galaxy)
  • Provides hardware-specific commands and configurations
  • Real-time telemetry monitoring (temperature, power, clock speed)
  • Multi-device support with aggregate health status

Interactive Learning Experience

  • Click-to-run commands from lesson content
  • Persistent terminal sessions maintain environment state
  • Visual progress tracking
  • Hierarchical organization by difficulty and category

Production-Ready Code

  • Tested templates for common workflows
  • Best practices from Tenstorrent engineering team
  • Scripts saved to ~/tt-scratchpad/ for easy customization
  • Hardware-specific optimization examples

Multi-Framework Coverage

FrameworkPurposeUse Case
vLLMProduction LLM servingOpenAI-compatible API, high throughput
TT-ForgePyTorch compilationMLIR-based experimental compiler
TT-XLAJAX/PyTorch XLAProduction compiler for JAX workflows
TT-MetaliumLow-level kernelsCustom ops and hardware programming

Hands-On Cookbook Projects

The Cookbook (Lesson 16) includes 5 interactive projects that run directly on Tenstorrent hardware:

Conway's Game of Life running on Tenstorrent hardware

Game of Life - Classic cellular automaton with TT-NN™ acceleration
View full animation →

 Particle Life simulation on Tenstorrent

Particle Life - Physics simulation with 10,000+ particles
View full animation →
Mandelbrot set fractal rendering

Mandelbrot Set - Fractal rendering with hardware acceleration

 Audio mel spectrogram processing

Audio Processing - Mel spectrogram computation

Plus: Image filters (blur, sharpen, edge detection) - all with complete source code and interactive tutorials.

Documentation

User Documentation

  • FAQ.md - Comprehensive troubleshooting (covers 90% of common issues)
  • Lesson Content - Interactive lessons accessible via the extension
  • CHANGELOG.md - Version history and release notes

Developer Documentation

Community & Support

Common Issues

"No hardware detected"

tt-smi -r      # Reset and rescan
sudo tt-smi    # Try with elevated permissions

See FAQ.md for complete diagnostic steps.

"ImportError: undefined symbol" (TT-XLA)

unset TT_METAL_HOME
unset TT_METAL_VERSION

TT-XLA requires clean environment. See Lesson 12 for details.

"vLLM won't start"

echo $TT_METAL_HOME    # Should be ~/tt-metal
echo $MESH_DEVICE      # Should match your hardware (e.g., N150)

See FAQ.md for systematic vLLM debugging.

For more troubleshooting, check the FAQ or join Discord.

Contributing

We welcome contributions! Here's how to get involved:

  1. Report Issues - Use our issue templates for bugs, content issues, feature requests, or new lesson ideas
  2. Improve Content - Lessons are in content/lessons/*.md - submit PRs for corrections or improvements
  3. Add Features - See CONTRIBUTING.md for development setup
  4. Validate Lessons - Test lessons on hardware and update metadata
  5. Join Discussions - Participate in Discord and GitHub Discussions

See CONTRIBUTING.md for:

  • Development setup instructions
  • Architecture and design principles
  • Code style and standards
  • Testing requirements
  • Pull request workflow
  • Packaging and distribution

Release Information

Latest Release: v0.0.401 (2026-04-23)

Highlights:

  • ⬡ Tensix Grid Visualizer — animated Canvas component embedded in lessons (VSCode + GH Pages); shows NOC routing, parallelism, kernel dispatch on real Wormhole™/Blackhole® chip grids
  • 🐍 ttlang-sim-lite — pure-Python, torch-free fork of the TT-Lang simulator; runs TT-Lang kernels in the browser via Pyodide with no hardware required
  • 🎮 Browser playground — write and run TT-Lang kernels client-side with Pyodide; kernels: eltwise_add, fused_mma, matmul_relu, matmul_1d
  • 🔬 Dev Container — devcontainer.json for simulator-only development; extension detects context and routes commands accordingly
  • ☁️ Cloud simulator API skeleton — FastAPI + WebSocket execution server with /sim-test PR comment trigger
  • 🔍 Drift detection scripts — check-sim-lite-drift.py (fork vs upstream), check-vendor-drift.py (all vendor repos: TT-Metalium, tt-vllm, TT-Inference-Server, tt-forge-models, ttsim)

Previous Release: v0.0.400 (2026-04-21)

Highlights:

  • Fixed assets/img/ being packaged twice in the .vsix bundle

See CHANGELOG.md for complete version history.

Version Support

VersionStatusNotes
0.0.x✅ CurrentActive development, full support

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

License Understanding

This software assists in programming Tenstorrent products. However, making, using, or selling hardware, models, or IP may require the license of rights (such as patent rights) from Tenstorrent or others. See LICENSE_understanding.txt for details.

Third-Party Licenses

This extension depends on several open source projects:

  • marked (MIT) - Markdown parsing
  • sanitize-html (MIT) - HTML sanitization
  • mermaid (MIT) - Diagram rendering

Run npm list --prod to see all production dependencies.

Acknowledgments

Built by the Tenstorrent community with contributions from:

  • Tenstorrent Developer Relations and Engineering teams
  • Open source contributors worldwide
  • Community members providing feedback and hardware validation

Special thanks to:

  • Beta testers who validated lessons on real hardware across all device types
  • Documentation contributors who improved clarity and caught errors
  • Bug reporters who helped us fix issues quickly
  • Community members suggesting new lessons and features

Tenstorrent Ecosystem:

Ready to start building AI on Tenstorrent hardware? Install the extension and open the Welcome page! 🚀

Questions? Check the FAQ or join our Discord community!