graphify

“A few weeks ago I started using graphify — if you haven't heard of it, it builds a knowledge graph of your entire codebase so your AI coding assistant actually understands the structure, not just the file it's currently looking at. Game changer for large projects. But I hit a problem fast. Every time Claude Code made changes — refactors, new files, updated logic — the graph went stale. Silently. No warning. Claude would keep answering questions based on a snapshot of the codebase from an hour ag”

“The TL;DR Most graph-based context tools have a fatal flaw: the agent ignores them. I benchmarked Graphify CLI (using suggestion hooks)and Graphify MCP (a registered Model Context Protocol server)against our tool, GrapeRoot MCP (where the agent cannot bypass retrieval). We ran them against the Medusa e-commerce repo (1,571 TypeScript files) across 5 real code-audit tasks using Claude Sonnet 4.5 via Claude Code CLI. The results: Both Graphify setups hit a 0% graph adoption rate. Claude completely”

“It's really bad on massive legacy java 8 codebases. Even when using it on subfolders with --update it produces a useless graph and all the nodes are named "community X". I've tried it multiple ways , even using sub agents to parallelize the construction of the graph and having them combine it into a single graph manually. Always got bad output. It seems to me that this tool should be perfect for massive legacy codebases , yet it only seems to work well on small to medium size newly made proj”

“When I heard about Graphify, I thought it was a dream come true. Less token usage, quicker responses, better understanding of the codebase, the works. I wanted to look into it and actually test it out to see what the actual difference is, these are the results: Graphify Which files handle authentication token verification? 33.2k - 80.7k Trace the complete execution path from where an HTTP request hits our API routes down to the raw database query or ORM call. List every intermediate class, middl”

“I came across this amazing GitHub repo - 100% open source and free. Graphify is a Claude Code skill that reads a folder of code, PDFs, markdown, and screenshots, builds a knowledge graph, and can also start an MCP stdio server. About 71x fewer tokens per query than reading the raw files. Post link / GitHub repo in comments below. Features /graphify on any folder: code, docs, papers, images Multimodal: Claude vision on screenshots, diagrams, whiteboard photos Interactive graph.html, Obsidian vaul”

“Spamware burned my session tokens in one go. Mods delete this spammer.”

“I entered the first prompt into Claude Code, and it used all my credits for that one task. I had to wait for four hours.”

“What’s the actual benefit of this? I’ve asked the AI against our codebase. It seems to think that the tools to scan code quickly already provide the means to build this relationship knowledge. And additionally the code still needs to read the actual lines of code anyways for certain scenarios. Where does this really provide value for the agent? Anyone have specific examples where it massively improved the AIs knowledge or speed/cost of deducing code knowledge?”

“Used both. Quick honest take, since they’re actually solving slightly different problems: CodeGraph is the more “always-on” option. Local SQLite + tree-sitter , runs as an MCP server with a native file watcher, so the graph re-syncs when you save a file. Fully local, no API keys, nothing leaves your machine, broad language support. Best fit if you want a fire-and-forget local index that just stays fresh. The catch: the index is machine-specific and gitignored, so it’s very much a per-devel”

“A few weeks ago I started using graphify — if you haven't heard of it, it builds a knowledge graph of your entire codebase so your AI coding assistant actually understands the structure, not just the file it's currently looking at. Game changer for large projects. But I hit a problem fast. Every time Claude Code made changes — refactors, new files, updated logic — the graph went stale. Silently. No warning. Claude would keep answering questions based on a snapshot of the codebase from an hour ag”

“So in theory understand-anything should still work fo local llms if your IDE uses a local model as a provider, right ? anyway when I last tried graphify it produced slop and didnt even work most of the time. Plus it couldnt scan my large legacy codebase because of the number of files. Like what is the point of using these tools on a small codebase anyway ? maybe that has changed since then, maybe someone in the comments can attest”

“Used it a lot. Found it to be very cumbersome to use. Constantly out of date. Complicated to setup and maintain. I switched to https://github.com/colbymchenry/codegraph and it’s worked a lot better for me. Updates the index itself automatically so much lower maintenance. Rust based so much faster than graphify. I have been happy with it after switching.”

“Used both. Quick honest take, since they’re actually solving slightly different problems: CodeGraph is the more “always-on” option. Local SQLite + tree-sitter , runs as an MCP server with a native file watcher, so the graph re-syncs when you save a file. Fully local, no API keys, nothing leaves your machine, broad language support. Best fit if you want a fire-and-forget local index that just stays fresh. The catch: the index is machine-specific and gitignored, so it’s very much a per-devel”

“Used it a lot. Found it to be very cumbersome to use. Constantly out of date. Complicated to setup and maintain. I switched to https://github.com/colbymchenry/codegraph and it’s worked a lot better for me. Updates the index itself automatically so much lower maintenance. Rust based so much faster than graphify. I have been happy with it after switching.”

“I’m not gonna lie, I don’t mean to offend but I have no idea why it’s gotten so popular. It seems fiddly, is a bit odd to set up, didn’t seem to save much time or improve outputs on a pretty large code base, and I don’t quite get why it’s any better than an LSP still My lack of understanding probably comes from my misunderstanding of what it is, but has anyone actually found this THAT crucial of a tool?”

“Okay so I ran the /graphify command. And it did what it does, but now I’m confused as to where and how to then open/view the graph itself”