How jitsi/skynet works
Main Components
• Core API Server: Represent the main application entry point (skynet/main.py) that orchestrates incoming API requests.
• STT (Speech-to-text) modules like streaming_whisper and vox.
• TTT modules (for assistant, openai_api, persistence, rag, summaries, etc.).
• Authentication services (skynet/auth – bearer, jwt, etc.).
• File watchers, monitoring tools, logging, and metrics.
• External Dependencies:
• Redis – used for caching or pub/sub as noted in the README.
• External Model Services – vLLM, Ollama, and possibly local ffmpeg for audio processing.
• Docker Infrastructure – the Dockerfile, compose files (compose-dev.yaml) and docker scripts indicate containerized deployment.
• Demos/Clients: The JavaScript client demo and the streaming-whisper demo (located in the demos/ folder).
• Supporting Scripts & Tools: Utility scripts like run.sh, Makefile, and CI/CD workflows that impact deployment and maintenance.
Relationships and Interactions
• Similarly, indicate interactions between the assistant modules and external AI model services (vLLM/Ollama) for summarization and decision making.
• How authentication components (JWT, bearer, customer ID) integrate by verifying and securing API requests.
• External services like Redis with the main server, indicating caching or state sharing.
• Docker and CI/CD pipelines (placed under .github/workflows and docker directories) as deployment/infrastructure layers which package the entire application.
Architectural Patterns and Technologies
• Modular/Plugin Architecture: Highlight that the system is built in a modular fashion so that specific functionalities (e.g., STT streaming, summarization) can be enabled/disabled as needed.
• Monolithic API Entry Point: Even though it’s modularized internally, the main server is a single API endpoint.
• Containerization: Indicate that Docker is used for deployment and development flows.
• Asynchronous Communication: Some modules (like streaming whisper) are built to handle websockets and asynchronous requests.
• Technologies & Frameworks: Note the use of Python (source files), Poetry for dependency management, Redis for caching, potentially GPU-enabled inference stacks (vLLM, Ollama), and ffmpeg for audio processing.
Specific Elements
• Use clear labels for each component. For example, “Core API Server”, “STT: Streaming Whisper”, “TTT: Assistant & Summaries”, “Authentication Service”, “Redis”, “Docker Infrastructure”, “Client Demo (JS)” etc.
• Rectangles for internal modules,