tinker

Thinking Machines Tinker — Cloud-to-Local AI pipeline.

Thinking Machines Tinker is how we train the small, focused AI models that power the / foxlight ecosystem. Instead of relying on one giant, slow model for everything, we train small "LoRA" adapters — each one really good at a specific job — in the cloud, then run them locally on your device. They're lightweight (typically 20MB–100MB), so we can swap them in and out on the fly without any cloud round-trip.