Workflow

Step 1: The user inputs a prompt into the Eclypse Syntax interface or selects from a set of pre-built units.

Step 2: The Orchestrator receives the prompt and prepares it for further action. It employs a Coordinating LLM to produce structured instructions based on the prompt, taking into account the context and user intent, and determines the appropriate unit for the request.

Step 3: Based on the instructions from the Coordinating LLM, the Orchestrator searches the local network directory for the corresponding Unit Schema. This schema provides details about the unit's identity, available actions, and relevant data sources.

Step 4: The Orchestrator then invokes the Unit Handler, passing along the structured instructions and the pertinent Unit Schema to the LLM.

Step 5: The Unit Handler interacts with necessary components or APIs to fulfill the user request, such as querying databases or external services. It also retrieves additional contextual data from the Vector DB, which contains model embeddings, to build a refined prompt for the LLM.

Step 6: The Unit Handler makes a call to the LLM with the enhanced prompt, Unit Schema, and additional context. The LLM processes this input and generates a response using its natural language understanding and generation capabilities, tailored to the user's query.

Step 7: The Orchestrator evaluates the response and determines if the Plugin Manager should be activated based on the prompt's processing.

Step 8: The Plugin Manager activates relevant plugins according to the LLM's response. These plugins may perform a range of tasks, such as fetching data from an oracle, executing compile instructions, or deploying code onchain.

Step 9: The Orchestrator assesses whether additional data is needed for the processing result. If so, it will repeat Steps 2-9 until the entire prompt has been addressed. Once complete, the Orchestrator consolidates the outputs from the LLM and the executed plugins to create a comprehensive response suitable for the user.

Step 10: After confirming that the response is coherent, relevant, and actionable, it delivers the final output to the user through the platform interface, completing the prompt processing cycle.