How tetherto/qvac works
QVAC is a local-first, cross-platform AI monorepo: its SDKs expose local model workloads, while its CLI can serve models through an OpenAI-compatible HTTP API. The principal documented workflow is loading a model, running inference, and returning streamed or complete results; the repository also contains many workload-specific engines and a RAG subsystem. Edges are limited to README statements and source-index references; much runtime wiring is unsampled, so uncertain links are omitted.
Main entry points: index.js Audio generation · server.ts Server lifecycle · HyperDBStorage.ts HyperDB storage · IngestionService.ts Document ingestion
Components 23
SDKs and provider
- TypeScript SDK
packages/sdk/src - Python SDK
packages/sdk-python/src/tetherto/qvac_sdk - AI SDK provider
packages/ai-sdk-provider/src/provider.ts
Local AI engines
- Speech recognition
packages/asr-ggml/src/engines - Audio generation
packages/audiogen-ggml/index.js - Audio formats
packages/audiogen-ggml/lib/audio-format.js - Audio backends
packages/audiogen-ggml/lib/backends.js - Audio models
packages/audiogen-ggml/models.js - Neural speech input
packages/bci-whispercpp/bci.js - BCI stream handling
packages/bci-whispercpp/lib/stream.js - OCR pipeline
packages/ocr-ggml/addon/src/model-interface/Pipeline.cpp
HTTP serving
- Serve CLI
packages/cli/src/serve/command.ts - Server lifecycle
packages/cli/src/serve/core/server.ts - OpenAI API routes
packages/cli/src/serve/extensions/openai/routes - Model loading
packages/cli/src/serve/core/load-manager.ts - Model catalog
packages/cli/src/serve/core/model-catalog.ts
RAG workflows
- RAG workspaces
packages/inference/src/rag/rag-workspace-manager.ts - Document ingestion
packages/rag/src/services/IngestionService.ts - Embedding service
packages/rag/src/services/core/EmbeddingService.ts - Embedding validation
packages/rag/src/schemas/embedding.ts - HyperDB storage
packages/rag/src/adapters/database/HyperDBStorage.ts - Document retrieval
packages/rag/src/services/RetrievalService.ts
Other components
- Developer or AI tool
Connections 17
- Developer or AI tool → TypeScript SDK: loads and runs
- Developer or AI tool → Python SDK: loads and runs
- Developer or AI tool → Serve CLI: starts server
- Developer or AI tool → OpenAI API routes: sends API request
- Audio generation → Audio models: uses models
packages/audiogen-ggml/index.js - Audio generation → Audio formats: uses formats
packages/audiogen-ggml/index.js - Audio generation → Audio backends: uses backends
packages/audiogen-ggml/index.js - Neural speech input → BCI stream handling: uses stream helpers
packages/bci-whispercpp/index.js - RAG workspaces → HyperDB storage: creates adapter
packages/inference/src/rag/rag-workspace-manager.ts - Document ingestion → Embedding service: generates embeddings
packages/rag/src/services/IngestionService.ts - Embedding service → Embedding validation: validates input
packages/rag/src/services/core/EmbeddingService.ts - Document ingestion → HyperDB storage: writes documents
- Document retrieval → HyperDB storage: queries documents
- Serve CLI → Server lifecycle: starts lifecycle
- Server lifecycle → Model loading: loads models
- Server lifecycle → OpenAI API routes: registers routes
- Model loading → Model catalog: resolves models