Desktop GUI Features¶
Last Updated: December 12, 2025
Purpose: Visual management for training, brains, chat, datasets, resources, and early Subbrains (experts) administration.
Note: Screenshots are planned for a future documentation update. For now, launch
aios guito explore the interface.
Status: Implemented core panels. Subbrains Manager is WIP (view and refresh registry, actions are placeholders).
Key Modules¶
- App bootstrap and panel wiring:
src/aios/gui/app/app_main.py,src/aios/gui/app/panel_setup.py,src/aios/gui/app/logging_setup.py - Services:
src/aios/gui/services/__init__.py,router.py,log_router.py - Panels:
- HRM Training:
src/aios/gui/components/hrm_training_panel/ - Brains:
src/aios/gui/components/brains_panel/ - Rich Chat:
src/aios/gui/components/rich_chat_panel/ - Datasets:
src/aios/gui/components/datasets_panel/ - Subbrains Manager (WIP):
src/aios/gui/components/subbrains_manager_panel/ - Resources:
src/aios/gui/components/resources_panel/ - Settings/Status:
src/aios/gui/components/*
Related Documentation¶
- Core training: CORE_TRAINING.md
- Dynamic Subbrains (MoE): DYNAMIC_SUBBRAINS_MOE.md
- Memory optimization: MEMORY_OPTIMIZATION.md
- Multi-GPU and parallel chunk system: MULTI_GPU_DISTRIBUTED.md, PARALLEL_TRAINING_BLOCK_CHUNK_SYSTEM.md
- GUI-specific multi-GPU behaviour: MULTIGPU_GUI_GUIDE.md
Panels and Flows¶
HRM Training Panel¶
- Create/train ACTv1 brains with presets, VRAM estimator, logs, and stop/resume.
- MoE awareness in estimator (shows total vs active params).
- Location:
src/aios/gui/components/hrm_training_panel/
Tips (Windows): - Prefer "Parallel Independent" mode for multi-GPU on Windows (DDP is CUDA-only and more finicky). See MULTI_GPU_DISTRIBUTED.md. - Use Optimize to search max context/batch fitting VRAM.
Brains Panel¶
- List brains (bundles), load/unload, view details, export/import.
- Scans
artifacts/brainsand temporary "router-generated" brains. - Location:
src/aios/gui/components/brains_panel/
Rich Chat Panel¶
- Chat with loaded brain, Markdown and code highlighting.
- Routing output formatter is shared service (
router.py). - Location:
src/aios/gui/components/rich_chat_panel/
Datasets Panel¶
- Discover/download datasets, verify metadata, manage cache.
- Location:
src/aios/gui/components/datasets_panel/
Resources Panel¶
- CPU/GPU utilization targets, throttle modes, real-time stats for training.
- Multi-GPU usage caveats and troubleshooting: see MULTIGPU_GUI_GUIDE.md
- Location:
src/aios/gui/components/resources_panel/
Subbrains Manager (WIP)¶
- View expert registry with counts (total/active/frozen), hierarchy, average routing weight, activations.
- Refresh reads
artifacts/experts/registry.json. Actions like create/delete/freeze and goal linking currently print "CLI command needed". Use CLI instead (see DYNAMIC_SUBBRAINS_MOE.md). - Location:
src/aios/gui/components/subbrains_manager_panel/
Debug Panel¶
- Centralized debug output with categorized logs via LogRouter; helpful during training/inference.
- Location:
src/aios/gui/app/logging_setup.pyregisters the log router; debug UI integrated in app.
Evaluation Panel¶
- Run evaluation suites and view results; compare runs.
- Location:
src/aios/gui/components/evaluation_panel/(if separated) or integrated evaluation view.
Tools & MCP Panel¶
- Manage Tools and MCP servers; configure and observe status.
- Location:
src/aios/gui/components/tools_panel/,src/aios/gui/components/mcp_panel/(module names may vary by integration status)
Settings and Themes¶
- Configure themes, fonts, logging level, cache, and other app settings.
- Location:
src/aios/gui/components/settings_panel/ - Available themes: Light Mode, Dark Mode, Matrix Mode, Halloween Mode, Barbie Mode
Running the GUI¶
From a configured environment, start the app:
If you need to pin a specific Python interpreter, ensure .venv is activated first.