Release v0.5.0

What's Changed

• fix(ui): call patrol tool with action=config instead of defunct agent.patrol_config by @rjckkkkk in #40
• fix(ui): darken explorer dropdown option list in dark theme by @xu16601526267 in #41
• fix: address onboarding and deployment metadata regressions by @xu16601526267 in #37
• catalog(models): qwen3-30b-a3b RDNA3.5 llamacpp-vulkan variant (AMD395 iGPU) by @rjckkkkk in #42
• fix: fallback when configured LLM model is unavailable by @xu16601526267 in #44
• catalog(models): enable Qwen3.6 tool calling defaults by @xu16601526267 in #45
• docs(readme): v2 ship — theatrical sections + visual anchors + β WebUI GIF by @rjckkkkk in #43
• docs(readme): Chinese README v2 mirror + commit spec source by @rjckkkkk in #46
• fix(ui): stabilize deployed model list rendering by @xu16601526267 in #49
• catalog(models): default AIBook Qwen variants to vllm-musa by @xu16601526267 in #47
• fix(model): catalog-drive AIBook preloaded model onboarding by @xu16601526267 in #48
• fix(model): harden catalog local registration by @skyguan92 in #52
• fix(catalog): honor local_path during deploy resolve by @xu16601526267 in #55
• fix(ui): add deployment stop and scan actions by @xu16601526267 in #53
• fix(ui): collapse component model section by default by @xu16601526267 in #54
• fix(ui): normalize unified memory display by @xu16601526267 in #56
• Discover and import external inference services by @rjckkkkk in #50
• Fix onboarding API access and AMD device support by @rjckkkkk in #51
• Fix deploy.delete compatibility for model-name stop actions by @xu16601526267 in #57
• fix: stabilize local model deployment state by @rjckkkkk in #59
• Fix cleared support active task status by @rjckkkkk in #69
• fix(catalog): point qwen3.6-35b-a3b text-only path at vllm-nightly by @rjckkkkk in #75
• fix(deploy): cap inference pod memory on unified-memory hosts by @rjckkkkk in #73
• Persist AIMA serve during onboarding init by @rjckkkkk in #70
• [codex] Split dry-run resolved and effective config by @xu16601526267 in #63
• [codex] Filter engine config flags by command context by @xu16601526267 in #64
• [codex] Expose service catalog context diagnostics by @xu16601526267 in #65
• fix(deploy): prefer a locally-cached engine image during auto-selection by @rjckkkkk in #74
• Fix OpenClaw sync recovery for stale deployments by @rjckkkkk in #71
• fix: sync agent and OpenClaw model routing by @rjckkkkk in #67
• [codex] Add deployment details review before starting models by @xu16601526267 in #66
• feat: wire running service details to real status by @xu16601526267 in #76
• [codex] Add deploy modal image preflight parity by @xu16601526267 in #77
• Detect Windows hardware via CIM, add AMD GPU detection by @rjckkkkk in #78
• Discover models across all drives; group split GGUF shards by @rjckkkkk in #79
• Discover off-PATH native engine binaries via AIMA_ENGINE_DIR by @rjckkkkk in #80
• Add Strix Halo llama.cpp GGUF model knowledge (aliases + verified perf) by @rjckkkkk in #81
• feat(model): mark speculative draft heads (DFlash/MTP) as non-standalone by @rjckkkkk in #82
• fix(external): don't list AIMA's own deployment backend (llama.cpp :8080) as an external service by @rjckkkkk in #83
• Add AMD ROCm/HIP Windows llama.cpp engine (auto-download on Strix Halo) by @rjckkkkk in #85
• Launch native engines from AIMA_ENGINE_DIR (scanned ⇒ launchable) by @rjckkkkk in #86
• Fix native deploy readiness: deploy.apply returned a name status lookup couldn't find by @rjckkkkk in #87
• Hide the deploy launcher console window on Windows by @rjckkkkk in #88
• Add Qwen2.5-VL-3B-Instruct model knowledge (vlm + verified perf) by @rjckkkkk in #89
• Auto-wire llama.cpp --mmproj for VL models (zero-config vision) by @rjckkkkk in #90
• openclaw sync: preflight-probe the AIMA proxy and warn loudly if unreachable by @rjckkkkk in #91
• feat(recovery): reconcile persistent deployment intent by @qinyue0508-cmd in #101
• feat(engine): manage verified engine lifecycle by @qinyue0508-cmd in #102
• fix(amd395): align Linux local model engines by @qinyue0508-cmd in #106
• fix(engine): activate compatible native versions by @rjckkkkk in #107
• fix: allow local native engines on current platform by @xu16601526267 in #97
• fix: refresh agent state after deployment by @xu16601526267 in #98
• feat: detect Houmo XH2A NPUs by @xu16601526267 in #99
• feat: add DeepSeek V4 Flash dual-Spark deployment by @xu16601526267 in #104
• release: v0.5.0 by @rjckkkkk in #110

Full Changelog: v0.4.0...v0.5.0