How absentdaemon-ctrl/kosmosxproject works
KOS-MOS is a desktop AI assistant whose primary workflow is voice input → speech recognition → LLM planning and tool calls → spoken response. Tool execution is mediated by safety checks, user confirmation, verification, and persistent task/audit records. The README documents launching the assistant and configuring Groq/HuggingFace credentials; sampled runtime wiring clarifies the voice, UI, tool, and storage paths. Some tool/API details and UI assets were not sampled, so those branches are represented at their supported directory level.
Main entry points: registry.py Tool Registry · pipeline.py Voice Pipeline · __main__.py Desktop App · loader.py MCP Tool Loader
Components 22
Desktop and Voice
- Desktop App
src/kos_mos/__main__.py - Voice Pipeline
src/kos_mos/voice/pipeline.py - Audio Capture
src/kos_mos/voice/audio_capture.py - Speech Recognition
src/kos_mos/voice/stt.py - Speech Output
src/kos_mos/voice/tts.py
Reasoning
- Groq LLM Adapter
src/kos_mos/llm/groq_adapter.py - Runtime Settings
src/kos_mos/config.py
Tool Execution
- Tool Registry
src/kos_mos/tools/registry.py - Computer Control
src/kos_mos/tools/computer.py - Desktop Tools
src/kos_mos/tools - Information APIs
src/kos_mos/tools/apis - MCP Tool Loader
src/kos_mos/mcp/loader.py - MCP Client
src/kos_mos/mcp/client.py - MCP Tool Adapter
src/kos_mos/mcp/tool.py
Safety and Verification
- Safety Policy
src/kos_mos/safety/policy.py - User Confirmation
src/kos_mos/safety/confirmation.py - Result Verification
src/kos_mos/verification/manager.py
Persistence and Telemetry
- Performance Profiler
src/kos_mos/profiler.py - Session and Tasks
src/kos_mos/storage/session.py - Audit Log
src/kos_mos/storage/audit.py - SQLite Store
src/kos_mos/storage/database.py
Other components
- User
Connections 28
- User → Desktop App: starts assistant
- Desktop App → Voice Pipeline: starts pipeline
src/kos_mos/__main__.py - Desktop App → User Confirmation: sets UI callbacks
src/kos_mos/__main__.py - User → Voice Pipeline: speaks requests
- Voice Pipeline → Audio Capture: captures audio
src/kos_mos/voice/pipeline.py - Voice Pipeline → Speech Recognition: transcribes speech
src/kos_mos/voice/pipeline.py - Voice Pipeline → Groq LLM Adapter: requests responses
src/kos_mos/voice/pipeline.py - Voice Pipeline → Tool Registry: dispatches calls
src/kos_mos/voice/pipeline.py - Voice Pipeline → Speech Output: speaks replies
src/kos_mos/voice/pipeline.py - Voice Pipeline → User Confirmation: cancels pending
src/kos_mos/voice/pipeline.py - Voice Pipeline → Session and Tasks: records completion
src/kos_mos/voice/pipeline.py - Voice Pipeline → Performance Profiler: tracks stages
src/kos_mos/voice/pipeline.py - Groq LLM Adapter → Runtime Settings: loads settings
src/kos_mos/llm/groq_adapter.py - Tool Registry → Safety Policy: evaluates risk
src/kos_mos/tools/registry.py - Tool Registry → User Confirmation: requests approval
src/kos_mos/tools/registry.py - Tool Registry → Result Verification: requests verification
src/kos_mos/tools/registry.py - Tool Registry → Audit Log: logs execution
src/kos_mos/tools/registry.py - Tool Registry → Session and Tasks: records execution
src/kos_mos/tools/registry.py - Tool Registry → Computer Control: registers tools
src/kos_mos/tools/__init__.py - Tool Registry → Desktop Tools: registers tools
src/kos_mos/tools/__init__.py - Tool Registry → Information APIs: registers tools
src/kos_mos/tools/__init__.py - MCP Tool Loader → MCP Client: connects server
src/kos_mos/mcp/loader.py - MCP Tool Loader → MCP Tool Adapter: wraps tools
src/kos_mos/mcp/loader.py - MCP Tool Loader → Tool Registry: registers tools
src/kos_mos/mcp/loader.py - MCP Tool Adapter → MCP Client: calls remote tool
src/kos_mos/mcp/tool.py - Session and Tasks → SQLite Store: reads and writes
src/kos_mos/storage/session.py - Audit Log → SQLite Store: appends events
src/kos_mos/storage/audit.py - Desktop App → MCP Tool Loader: loads MCP tools
src/kos_mos/__main__.py