You can now run a 105GB AI model on a Mac that doesn't have enough RAM for it.
slotstream streams the model from your SSD and keeps only what's needed in memory. It runs Qwen3.8-Flash-Next (125B parameters) on Macs with 16-64GB RAM. Fully offline. No cloud account needed.
On a 48GB M5 Pro: ~21-24 tokens/sec.
You can even point Claude Code at it:
slotstream launch claude
Codex, opencode, and Hermes work the same way.
This is huge for anyone running local AI without enterprise-grade hardware. No more RAM bottlenecks.
slotstream streams the model from your SSD and keeps only what's needed in memory. It runs Qwen3.8-Flash-Next (125B parameters) on Macs with 16-64GB RAM. Fully offline. No cloud account needed.
On a 48GB M5 Pro: ~21-24 tokens/sec.
You can even point Claude Code at it:
slotstream launch claude
Codex, opencode, and Hermes work the same way.
This is huge for anyone running local AI without enterprise-grade hardware. No more RAM bottlenecks.