Token reduction
Graphify glossary
The context-window savings from answering with graph structure instead of pasting whole files into an assistant's context.
Token reduction is the context-window savings you get when an assistant answers from a graph instead of from raw files. The two costs being compared: when an assistant greps, it opens the files that match and pastes them into context wholesale — every import block, every unrelated function, every license header — and pays for every token, on every question, in every session. When it queries a graph, it receives just the relevant nodes and edges: the callers of the function, not the full text of every file that mentions its name.
The savings compound for two reasons. First, structural answers are small — 'these 12 functions call processPayment, here are their locations' is a few hundred tokens, where the grep-and-read loop behind the same answer might pull in dozens of files. Second, the graph persists between sessions, so the assistant isn't re-deriving the architecture from scratch each time; the mapping cost is paid once at /graphify . time, then amortized over every question after.
Honest framing matters here: actual savings depend entirely on your repo, your questions, and how your assistant explores. One community user reported 71.5× fewer tokens on their workload — a community-reported figure from real usage, not a controlled benchmark, and your number will differ. The durable claim is directional: traversing structure sends less text to the model than pasting files, which means lower cost and more of the context window left for the actual task.
Related terms
- Traversal / hop — Following edges from node to node — one hop is a single edge, a traversal is a path of them. How graphs answer multi-step questions.
- RAG (retrieval-augmented generation) — A pipeline that chunks documents, embeds them as vectors, and retrieves the most similar chunks at query time — lookup by resemblance, not structure.
- MCP server — A server implementing the Model Context Protocol — the open standard that lets AI assistants call external tools, like a code graph.