v0.56.0

What's Changed

• Disable windows by @i386 in #143
• Remove client nodes from topology visualization by @i386 in #145
• feature: mesh topology selector by @ndizazzo in #152
• add navigable node detail sidebar by @i386 in #148
• [codex] make README friendlier and split long reference docs by @i386 in #155
• Add workflow to clean up PR GitHub Actions caches on close by @i386 in #158
• Add GitHub cache versioning and a full cache reset workflow by @i386 in #159
• update hf-hub to checksum verification rev by @i386 in #154
• Update mesh-llm crate README to match current architecture by @i386 in #157
• Add explicit update command and opt-in auto-update flow by @i386 in #138
• fix: --no-draft flag ignored + case-sensitive cache path mismatch by @vveerrgg in #160
• Add multimodal blob uploads and console support by @i386 in #153
• Deflake CI smoke tests by @michaelneale in #167
• Fix llama-server failing to start on single-node, update llama.cpp fork to upstream by @michaelneale in #172
• [codex] Namespace server logs by mesh-llm PID by @i386 in #173
• test(ui): add selective PR-130 regressions and HF cache matching by @ndizazzo in #162
• [codex] Sync owner keystore crypto with main by @i386 in #178
• crypto: owner keystore and signed+encrypted message primitives by @Bortlesboat in #156
• feat: auto-enable row split for local multi-GPU (tensor parallelism) by @michaelneale in #166
• Build the new plugin system for MCP, HTTP, and inference endpoints by @i386 in #169
• Resolve managed models from the Hugging Face cache only by @i386 in #175
• Add serve/client commands and local GPU inspection for #184 by @i386 in #185
• local config support on top of #185 by @i386 in #186
• Automatic MoE expert ranking improvements by @i386 in #137
• [codex] Document Lemonade integration by @i386 in #190
• Use OS keychain for storing identity keystore by @i386 in #187
• [codex] Keep /api/status off model inventory by @i386 in #192
• feature + fix: fixed layout bug, node selection tooltip bug, added table selection by @ndizazzo in #198
• Remove deprecated models migrate command by @i386 in #200
• fix: hard crash when splitting KV quantizations with flash attention by @ndizazzo in #199
• feat(protocol): add distributed config protocol with ownership by @ndizazzo in #193
• Make release version bump fail fast and update all tracked manifests by @i386 in #201

New Contributors

• @vveerrgg made their first contribution in #160
• @Bortlesboat made their first contribution in #156

Full Changelog: v0.55.1...v0.56.0