v0.64.0

What's Changed

• fix: remove unintended PR job from 307 by @ndizazzo in #310
• Share split GGUF MoE rankings under stable distribution refs by @i386 in #314
• feat: Implement light mode for topology diagram by @ndizazzo in #302
• [codex] Bump llama.cpp fork pin to e88186e78777 by @i386 in #315
• Use published MoE analysis for fit preflight by @i386 in #316
• Update LLAMA_CPP_SHA hash value by @i386 in #320
• models: inspect and clean up mesh-managed cache entries safely by @IvGolovach in #300
• Align MoE split cache naming with llama.cpp by @i386 in #321
• Run releases from GitHub Actions instead of tag pushes by @i386 in #323
• feature: add embedded SDK support by @ndizazzo in #233
• Let moe analyze share rankings automatically by @i386 in #325
• Add optional CUDA and ROCm skip for prerelease releases by @i386 in #326
• Keep auto-routed chats on one model and one peer by @michaelneale in #324
• Codex/catalyst xcframework by @i386 in #329
• Stop small models from looping in decode by @michaelneale in #328
• Show GGUF variants on show and variant counts in search results by @i386 in #332
• Auto routing prefers larger models when they are available by @michaelneale in #334
• Rename PR github copilot instructions by @Michionlion in #338
• fix(light-mode): the toplogy view in light mode was not readable by @ndizazzo in #333
• fix(ui-tests): add missing UI tests for React to CI and Justfile by @ndizazzo in #340
• Stop floods from piling up inside llama-server by @michaelneale in #341
• Add reusable command bar / filter bars, ship the first model catalog filters by @ndizazzo in #306
• Fix client join hang on unresponsive gossip peer by @michaelneale in #346
• Richer image understanding and multi-file attachments by @michaelneale in #336
• fix(ui): fix the unintended blue line in dark mode by @ndizazzo in #347
• status: surface routing outcomes and utilization in the management API by @IvGolovach in #301
• feature: revise node states to be simpler by @ndizazzo in #298
• Add --headless flag to disable the web UI while keeping the management API alive by @ndizazzo in #349
• Fix broken tests on main: add routing_metrics to StatusPayload test literals by @michaelneale in #352
• fix(ui): improve accessibility and code quality in UI components by @JiayangLai in #342
• Fix: Update AMD device naming from HIP to ROCm to match upstream llama.cpp by @intenso in #319
• auto: only join community mesh, never private named meshes by @michaelneale in #354

New Contributors

• @Michionlion made their first contribution in #338
• @JiayangLai made their first contribution in #342
• @intenso made their first contribution in #319

Full Changelog: v0.62.1...v0.64.0