Welcome to Your TT-QuietBox 2
Welcome to Your TT-QuietBox 2
Four Blackhole chips · 480 Tensix cores · 128 GB GDDR6 · no API keys required
From open source models to local agents to custom kernels and infinite generative art. Welcome to your TT-QuietBox 2.
Chat with AI
Private LLM inference — 32B & 70B scale
Start here
TT-Studio — browser UI, one-click model deploy, pre-installed
Models
Llama-3.3-70B ~14s/response · Llama-3.1-8B fast · full catalog ↗
API
TT-Inference-Server — OpenAI-compatible endpoint, pre-installed
Monitor
TT-Toplike — watch cores during inference
Generate Video & Images
Text-to-video, image-to-video, stills
Start here
TT-Inference-Server — deploy and serve generation models, pre-installed
Gallery
TT-Local-Generator — queue, gallery, TT-TV kiosk mode
Video
Wan 2.2-14B ~6 min/5-sec clip
Image
FLUX.1-dev high-quality stills
Prompts
✨ Inspire me — prompt ideas from a small on-device model
Build AI Agents
Tool calling, multi-agent, stateful pipelines
Best model
Llama-3.3-70B — best for agents
Frameworks
smolagents · CrewAI · OpenAI Agents SDK
Demos
Web research · codebase nav · multi-role pipelines · stateful dungeon master
Privacy
Data never leaves the machine — architecture, not policy
Hack the Hardware
There’s a real computer in here — boot Linux, write kernels
Boot
tt-bh-linux — run Linux on 16 RISC-V cores inside the chip
Demos
Conway’s Game of Life · Zork on TT hardware
Kernels
tt-lang — Python DSL for Tensix kernels · TT-Metalium — C++ dispatch
Explore
tt-tiny — George Hotz’s minimal bare-metal Python exploration
Explore the Architecture
TT-Toplike, Particle Life, TT-Metalium
Watch
TT-Toplike — Starfield, Memory Castle, Memory Flow, Arcade
Run
Particle Life Simulator on 4× chips — emergent complexity
Compile
TT-Forge — bring any PyTorch or ONNX model to the hardware, 100+ supported
Program
TT-Metalium — C++ kernels on RISC-V cores in the 2D mesh
Discover
TT-Awesome — community apps, demos, libraries, and write-ups
Lessons
CS Fundamentals series · Particle Life walkthrough
Train
TT-Blacksmith — fine-tune or train from scratch on 4× Blackhole
| Tool | What it does | Where |
|---|---|---|
| TT-SMI | Hardware status and telemetry snapshot | Pre-installed |
| TT-Studio | Web GUI for deploying and chatting with AI models — handles all setup automatically, deploy models in one click. Gated models (e.g. Llama) require a Hugging Face token. | Pre-installed — run tt-studio |
| TT-Inference-Server | OpenAI-compatible model serving endpoint | Pre-installed at ~/.local/lib/tt-inference-server |
| TT-Toplike | Real-time hardware visualization — power, temperature, core activity | docs.tenstorrent.com/tt-toplike |
| TT-Local-Generator | Local video and image generation queue and gallery | docs.tenstorrent.com/tt-local-generator |
| TT-VSCode-Toolkit | Guided lessons and architecture walkthroughs, validated on QB2 | docs.tenstorrent.com/tt-vscode-toolkit |
| TT-Forge | MLIR compiler frontend — run PyTorch, ONNX, and JAX models on QB2 hardware, 100+ models supported | docs.tenstorrent.com/tt-forge |
| TT-Metalium | Low-level Tensix programming — C++ kernels on RISC-V cores | docs.tenstorrent.com |
| TT-Blacksmith | Optimized training recipes — fine-tune and train models from scratch on QB2 | docs.tenstorrent.com/tt-blacksmith |