Bird's Nest launches non-transformer local AI

This is a smart niche bet: instead of trying to out-Ollama Ollama, Bird's Nest builds a home for architectures most local AI tools barely acknowledge.

• –The pitch is genuinely differentiated because it centers non-transformer inference, especially RWKV and Mamba, rather than treating them as side experiments
• –Constant-memory RNN-style inference is the real hook for developers who care more about long-running local sessions than benchmark leaderboard bragging rights
• –Shipping 25+ built-in tools plus local image and music generation makes it feel more like a self-hosted AI workstation than a bare model runner
• –The downside is obvious too: macOS-only support and a very early 0.1.0 release mean this is still a power-user project, not a polished mainstream local AI platform
• –If non-transformer models keep improving, Bird's Nest could become the default control plane for a category the current local inference stack underserves