Atomic Chat v1.1.76

🚀 New Features

• Multi-Token Prediction (MTP) for llama.cpp on macOS — speculative multi-token decoding for faster inference on macOS.

• Reasoning context management — chain-of-thought context is now tracked automatically for local reasoning models.

• Updated llama.cpp to latest upstream — kernel optimizations and model format fixes from upstream.

🔧 Improvements

• Removed legacy cache_type_k / cache_type_v overrides — conflicted with TurboQuant. Provider label updated.

• Local API now binds to 127.0.0.1 by default — set host: 0.0.0.0 to expose on LAN.

• Goose and nanobot added to "Launch With" docs — connect them to the local API on port 1337. Config snippets in the docs.

🙏 Contributors

Big thanks to @Vect0rM, @yanalialiuk, and @Fieldnote-Echo — this release wouldn't have shipped without them 🫡