How metil-holoai/holotable-middleware works
Project type and purpose
• This is a Go-based middleware service (“HoloTable-Middleware”) that sits between various input sources (text, microphone, speech-to-text), a set of AI content-generation workflows (GIF, image, video, 3D, intent detection, etc.), a database, and an external Unity application.
• It exposes REST and WebSocket endpoints for real-time or on-demand content requests, orchestrates prompts and workflows defined in YAML, and handles storage and retrieval of generated assets.
File structure
• Top-level directories:
• cmd/holoai: the entrypoint (main.go)
• config: all YAML workflow definitions and general configuration
• internal: core code, subdivided into:
– callers: orchestration of content extraction, generation, storage, prompt loading
– config: environment loading and struct definitions
– database: persistence layer
– endpoints: HTTP REST API and WebSocket server
– listeners: adapters for text, microphone, Vosk speech recognition
– unityserver: export logic and structs for Unity integration
– utils: error checking and shutdown handling
• Go.mod/go.sum define dependencies.
• .github/workflows contains CI build test definition.
Extract insights from README
• Minimal top-level README just points to “Run the Middleware” and “Add a Workflow” docs, confirming that the core functionality is dynamic workflow orchestration.
Elements
• Adapters (listeners, endpoints)
• Core logic (callers)
• Infrastructure (database, external services)
• YAML config files alongside the configuration loader component
Final advice
• Be as detailed as possible: break out each listener type, each caller sub-component, and each external model if it helps clarity.
• This approach ensures any new engineer or stakeholder immediately understands how input enters the system, how workflows are executed, and how results are delivered or stored.