A node processor for @Spounge that manages execution context via an MCP server and proto contracts, supporting LLM calls, API integrations, database operations, and LangGraph workflow execution. Requires Polykey for secure credential fetching and granting.

Structural Insight & Execution Flow

Nodus employs a layered architecture for workflow processing:

1. Executor Layer (executors/): Executes specific node tasks (Direct, Reasoning, Autonomous, Webhook). Executors are isolated from external service interactions and high-level decision-making.
2. Controller Layer (controllers/): Manages orchestration and decision logic. For example, LLMManager determines required LLM models or capabilities. This layer delegates external interactions.
3. MCP Layer (mcp/): Exposes external tools and services. The MCPClient provides an interface for dynamic tool discovery and invocation, centralizing access.
4. Integration Layer (integrations/): Contains thin adapters for direct communication with third-party services (APIs, SDKs, LLMs, external infrastructure like GeminiLLMProvider, PolykeyClient). This layer encapsulates service mechanics without business logic.

Execution Flow: An incoming request is routed to a designated executor. If the executor requires external capabilities (e.g., LLM inference or tool execution), it forwards the request to a controller (e.g., LLMManager) or directly to the MCPClient. The controller/MCP then utilizes the integration layer to interact with the external service.

Current Status

Polykey integration: Awaiting full implementation for secure credential fetching and granting.