Transformers.js
In-process AI via WASM. No server, no setup, fully offline.
Why Transformers.js
Transformers.js is the simplest way to run AI in Wilson. It loads models directly into the Bun process via WebAssembly — no Ollama install, no Docker, no server management. Just install Wilson and go.
Models download automatically on first use and cache locally. After the initial download, everything runs fully offline. No network calls, no API keys, no accounts.
The trade-off is smaller model selection and slower inference on CPU compared to Ollama. But for basic categorization and embeddings, it's more than enough — and the zero-setup experience is hard to beat.
What you get
Zero config — models download automatically on first use
In-process via Bun — no external server to install or manage
WebGPU support for hardware-accelerated inference
Fully offline after initial model download
Supported models
Models run via ONNX Runtime. WebGPU models require a compatible GPU and Bun build with WebGPU support.
Model
Runtime
Best for
Xenova/qwen2.5-0.5bXenova/all-MiniLM-L6-v2onnx-community/Qwen2.5-1.5Bonnx-community/Phi-3.5-miniOllama vs Transformers.js
Ollama
Transformers.js
Setup
Install Ollama + pull model
Zero — built in
Model size
2–7B+ parameters
0.5–1.5B parameters
Speed
Fast (GPU-accelerated)
Slower (WASM/WebGPU)
Accuracy
Higher
Good for basics
External server
Yes (localhost)
No — in-process
Best for
Daily use, complex analysis
Quick start, simple tasks
Zero-setup AI.
Install Wilson and start analyzing. Transformers.js handles the rest.