How bhoomik-codes/kurumi works
KURUMI is a local-first AI application whose persistent daemon serves inference, storage, and worker operations to CLI, TUI, and Electron clients. The main workflow is a user query through a client to the daemon, which streams a selected provider’s response and persists conversation history. The GUI also exposes model management, document RAG, image generation, voice, and interactive artifacts. The README documents the daemon-client relationships and capabilities; sampled source confirms key daemon, worker, and Electron wiring, but not every GUI subsystem’s implementation.
Main entry points: server.ts HTTP daemon · App.tsx GUI routes · Chat.tsx Chat interface · workerManager.ts Worker manager
Components 24
Client surfaces
- CLI client
src/cli/index.ts - Terminal UI
src/cli/tui.tsx - Electron host
electron/main.ts
Daemon core
- HTTP daemon
src/daemon/server.ts - Conversation database
src/daemon/db.ts - Process supervisor
src/daemon/supervisor.ts
Inference providers
- Ollama provider
src/daemon/services/OllamaService.ts - AirLLM provider
src/daemon/services/AirLLMService.ts - NVIDIA provider
src/daemon/services/NvidiaService.ts
Knowledge and voice
- Worker manager
src/daemon/workerManager.ts - RAG worker tasks
src/daemon/worker/ragWorkerTasks.ts - Vector store
src/daemon/services/vectorStoreCore.ts - Voice transcription
src/daemon/worker/voiceWorkerTasks.ts
GUI capabilities
- Chat interface
src/pages/Chat.tsx - GUI routes
src/App.tsx - Document management
src/pages/Documents.tsx - Image generation
src/pages/ImageGen.tsx - Model management
src/pages/Models.tsx - Model store
src/pages/ModelStore.tsx - Interactive artifacts
src/components/artifacts/ReactArtifact.tsx
Other components
- User
- Ollama runtime
- AirLLM server
airllm_server.py - NVIDIA API
Connections 29
- User → CLI client: queries
README.md - User → Terminal UI: chats
README.md - User → Chat interface: chats
README.md - CLI client → HTTP daemon: HTTP requests
README.md - Terminal UI → HTTP daemon: HTTP requests
README.md - Electron host → HTTP daemon: starts and checks
electron/main.ts - Chat interface → HTTP daemon: streams chat
README.md - GUI routes → Chat interface: routes to
src/App.tsx - GUI routes → Document management: routes to
src/App.tsx - GUI routes → Image generation: routes to
src/App.tsx - GUI routes → Model management: routes to
src/App.tsx - GUI routes → Model store: routes to
src/App.tsx - Chat interface → Interactive artifacts: renders
- HTTP daemon → Conversation database: reads and writes
src/daemon/server.ts - HTTP daemon → Ollama provider: dispatches inference
src/daemon/server.ts - HTTP daemon → AirLLM provider: dispatches inference
src/daemon/server.ts - HTTP daemon → NVIDIA provider: dispatches inference
src/daemon/server.ts - Ollama provider → Ollama runtime: calls
- AirLLM provider → AirLLM server: calls
src/daemon/services/AirLLMService.ts - NVIDIA provider → NVIDIA API: calls
- HTTP daemon → Process supervisor: starts processes
src/daemon/server.ts - HTTP daemon → Worker manager: routes worker jobs
README.md - Worker manager → RAG worker tasks: dispatches tasks
- RAG worker tasks → Vector store: indexes and searches
src/daemon/worker/ragWorkerTasks.ts - Document management → HTTP daemon: uses RAG API
- Worker manager → Voice transcription: dispatches transcription
- Image generation → HTTP daemon: uses image service
- Model management → HTTP daemon: manages models
- Model store → HTTP daemon: uses model API