Why We Built an AI-Native Learning Map
Conversational AI is brilliant at answering isolated questions, yet linear chat threads lose sight of prerequisite structure. Here is why Consyllus pairs Directed Acyclic Knowledge Graphs with MCP.
When studying a rigorous subject like Linear Algebra, Probability Theory, or Real Analysis with an AI assistant, a familiar problem quickly emerges: chat threads are linear and ephemeral, whereas real knowledge is a Directed Acyclic Graph (DAG).
The Failure Mode of Unstructured AI Tutoring
Ask a language model to explain the Spectral Theorem before you have mastered Orthonormal Bases or Invariant Subspaces, and it will gladly produce a fluent explanation filled with concepts you cannot yet verify. Three prompts later, the chat history has drifted into a maze of tangents, and neither you nor the AI remembers which prerequisites are solid and which are still unexamined black boxes.
Separating Objective Topology from Personal Progress
Consyllus resolves this by splitting every subject into three explicit local files inside your Vault:
graph.json(Immutable Objective Topology): Every node declares the exact concepts itteachesandrequires. Prerequisite edges are deterministically projected from these concept links, verified to be cycle-free via Kahn’s algorithm, and sealed with an RFC 8785 SHA-256 content hash.outline.json(Two-Level Textbook Mapping): Maps authoritative textbook chapters and sections directly to graph nodes so you never lose your place in the source material.progress.json(Strictly Local Personal State): Tracks your completed nodes, active study target, and error logs on your own disk—never uploaded to any server.
Closing the Loop with the Model Context Protocol (MCP)
Instead of pasting your syllabus and study notes into every new conversation, Consyllus includes a zero-dependency local MCP server. When you study with Claude, Cursor, or Antigravity, your assistant queries your live topological boundary:
- Which foundational concepts are already mastered;
- Which node is your current study target;
- Which advanced terms should be treated as explicit blackbox terms until you reach their target chapters.
When you finish a node and verify comprehension, checking it off cascades across the DAG—unlocking downstream nodes whose prerequisites are now complete.