How leofroeder-stack/aria-decision-engine works
aria-decision-engine accepts text, a plain-language output schema, and optional few-shot examples, then returns a structured classification with latency and token metadata. Users can invoke it through the command-line interface or import decide() as a JavaScript module; MCP-compatible agents can call the same function through the stdio MCP server. The decision function sends a constrained JSON-mode request to Groq and parses the response. README examples document triage, urgency, spam, content filtering, and DTC severity use cases; example implementations are not mapped here. Runtime relationships are based on the sampled entry-point sources.
Main entry points: index.js Decision function · mcp-server.js MCP tool server
Components 9
Invocation interfaces
- CLI entry
index.js - Module API
index.js - MCP tool server
mcp-server.js
Decision runtime
- Decision function
index.js - Decision result
index.js
Other components
- CLI user
- JS caller
- MCP client
- Groq API
Connections 13
- CLI user → CLI entry: provides input
- CLI entry → Decision function: calls
- JS caller → Module API: calls
- Module API → Decision function: invokes
- MCP client → MCP tool server: calls tool
- MCP tool server → Decision function: invokes
- Decision function → Groq API: sends request
- Groq API → Decision function: returns response
- Decision function → Decision result: parses and measures
- Decision result → CLI entry: prints JSON
- Decision result → Module API: returns
- Decision result → MCP tool server: returns
- MCP tool server → MCP client: sends result