Shourya_55

Hi HN community , I am working on knowledge graph engine for codebases that basically converts your entire static codebase into a dynamic indexed knowledge graph represented in the form of Nodes .

Some of the features : 1) Identifying blast radius before any code change 2) Deranks and flags stale files internally 3) Uses less tokens per task rather than Grep by 96% to 99% tokens compared to standard grep 4)Integrates with claude code , codex , opencode etc 5) works both hosted/on device(sovereign)

I am open to any queries in comments

Although accuracy and token savings are independent measurements Novgraph achieved 85.7% accuracy across seven evaluated tool results, while measurable retrieval responses were 96.04%–99.81% saving tokens.

Sure

Glad you got it , Would you be open to use the product at trynovaya.com ?

I would be eager to know your feedback as well

API as a source of distribution is much more scalable and comfortable as well especially if user is actively engaging with multiple different Agents Eg you use Claude code ,codex or grok build for a single repository .

Most open source tools parse the code into graph even open source tools with 100k stars do the same AST parsing only.

We record why each change was made, written back by the agents as they work, so the graph gets better the longer you use it . You can even check out the dashboard that is always in synch with Commits and updates from GitHub and Gitlab .

We especially flag the stale files and code and derank them in internal . So, Graph stays in synch and updated .

Token savings are in range of 78 to 99.74 percent .

API as a source of distribution is much more scalable and comfortable as well especially if user is actively engaging with multiple different Agents Eg you use Claude code ,codex or grok build for a single repository .