Slotstream

Slotstream runs Qwen3.8-Flash-Next on Apple Silicon Macs with too little RAM to hold the whole model. It's aimed at Macs with 16 to 64 GB of memory, for people who want local chat, answers about pictures or a backend for coding assistants.

• Models larger than RAM: Most model weights stay on the SSD while frequently used parts stay in memory, keeping the full model available without loading it all into RAM.
• Chat and coding: You can ask text and image questions or connect Claude Code, Codex, Pi, opencode and Hermes to the local model.
• App connections: Ollama-, OpenAI- and Anthropic-compatible APIs let other apps use the engine, and a Swift library supports integration into native apps.
• Memory and conversation reuse: The engine adjusts its cache as other apps need memory and reuses unchanged conversation history to reduce repeated prompt processing.

Slotstream requires Apple Silicon, macOS 14 or later, at least 16 GB of RAM and about 110 GB of free SSD space. Windows, Linux and Intel Macs aren't supported. The native Swift engine uses Apple's MLX and Metal without Python. After the model download, inference works offline and doesn't require a cloud account. The engine is open source under the MIT license.

Long prompts can take minutes to process before a reply starts. Speed depends on the chip, SSD and available memory. Slotstream also runs on Macs with 96 GB or more, but engines that keep the whole model in RAM may reply faster on those machines.