How does this differ with graphify?

[DenTheProgrammer]

https://github.com/safishamsi/graphify - this
Seems very similar, but I can't decide which is better for me :)

[DeusData]

Hey, good question. The overlap is real: both tools try to give coding agents a graph-shaped view of a project instead of relying only on grep/read loops.

The most direct comparison I have seen so far is the open community benchmark PR here: #628. It is still an open PR, so I would treat it as community-supplied evidence rather than an official project claim, but the results are useful.

In that benchmark, run on this repo's C core, both tools were very strong at finding function definitions: Graphify reported 0.999 recall and CBM 0.997. The differences showed up more in graph richness and operational behavior:

• CBM had a richer typed schema: 11 node kinds vs Graphify's flatter code-node representation in that benchmark.
• CBM found more include/dependency edges: 326 vs 68.
• CBM scored higher on the Q&A evaluation: 3.48 vs 2.60.
• CBM was faster on cold build: 2.44s vs 4.21s.
• CBM had lower median query latency: about 40ms vs 126ms.
• The report also notes that Graphify's LLM/community-labeling path is more useful for report/label generation and mixed corpora, but did not improve pure-code structural accuracy in that C-code benchmark.

So my short take would be:

Graphify is broader: code, docs, PDFs, images, videos, visual graph/report output, and a general project knowledge graph workflow.

CBM is narrower and more code-specialized: local MCP server, structural code search, exact snippets, call tracing, impact analysis, architecture queries, and low-latency graph operations for coding agents.

One important distinction is product direction: CBM is built primarily for coding agents, not for humans browsing a graph. The graph quality is the product surface. We are actively working on improving that graph, especially through more and better Hybrid LSP layers, because better symbol resolution, call edges, type information, and cross-file relationships directly improve agentic coding performance. The goal is to give the agent the most accurate, low-latency structural context possible so it can work with fewer tokens, fewer wrong turns, and better code-change decisions.

If your corpus is mixed docs/media plus code, Graphify may be attractive. If your main use case is code-only agent navigation and structural code questions, the direct comparison data I have seen points toward CBM. CBM's direction is deliberately agent-first: the coding agent is the primary user, and better graph context is the lever for better agent efficiency.

[DenTheProgrammer]

Thanks for the detailed answer!