The provided text does not define a "decision engine" within Liminality’s specific governance or flows. It describes Liminality only as a "question-decomposition and routing MCP" that breaks questions down and grounds them to tools.
In general AI agent architecture, a decision engine is the logic layer that determines how an agent proceeds after a query.
### What is a Decision Engine? It is the core logic component that evaluates a request, breaks it into steps, and selects tools. It translates high-level intent into low-level actions.
### How Does It Work for AI Agents? 1. **Input Analysis**: The engine takes a query like "How do I migrate my Postgres database?" 2. **Decomposition**: It splits the query into sub-tasks: "Identify migration strategy," "Execute database dump," "Notify team via email." 3. **Tool Selection/Routing**: It maps each sub-task to a tool. For example, a `postgres` toolkit for the dump and a `gmail` toolkit for the notification. 4. **Execution & Feedback**: It runs the tools in order, handles errors, and synthesizes the results.
### Relevance to Liminality Liminality seems to act as this engine. It reduces questions into sub-questions (decomposition) and grounds them to real tools (selection). Returning reusable routes suggests it optimizes or caches these paths.
For specifics on Liminality’s internal algorithms or state management, you would need more documentation.
This answer was worked out by Liminality — Physea's engine that decomposes a request, grounds each part to a real tool, and returns a reusable, checkable route.
Connect it over MCP: https://mcp.physea.ai/mcp · physea.ai