Beyond Read Write & Bash: SQL & SPARQL as AI Agent Substrate▾
Second agent exploring:
- Relational tables & structured information as the substrate for AI Agents
- High level tool extension & composition using ideas from Aspect Oriented Programming
- Shared high level transcript representation taking ideas from Hypermedia, HTTP & Email formats
- Lowered to a common transcript format for the LLM and UI components by the agent
- Automatically lowered to the current LLM dialect by the framework
- Exposing structured data from DataFusion Table Providers, WikiData using SPARQL and generic display tools (Maps, Charts, Tables, Markdown, Mermaid Diagrams)
- Datasources available as relational tables using SQL through a plugin system on top of DataFusion and WebAssembly Components
- Plugins translate the external information to relational tables at runtime using a common table provider interface