Display Name
Knowledge Graph
Category
Tooling
Sub-Category
General
Primary Link
https://github.com/hilyfux/knowledge-graph
Author Name
hilyfux
Author Link
https://github.com/hilyfux
License
MIT
Other License
No response
Description
Built on Anthropic internal engineering practices and Karpathy's AutoResearch methodology. A zero-dependency, git-native memory layer for Claude Code that persists learned context across sessions. Pure Bash with ~3ms/event overhead, privacy-first.
Validate Claims
Install the tool in a small repo, start a fresh Claude Code session, perform a few edits, then start a second session and ask Claude to recover prior file-level learnings and decisions from the generated project memory. This demonstrates persistent context across session boundaries without a database.
Specific Task(s)
In a test repo, make a few changes across files in session one, then open session two and ask Claude to summarize the previous session's learned conventions and touched files from the generated knowledge artifacts.
Specific Prompt(s)
After installing Knowledge Graph, make a few edits, end the session, start a new Claude Code session, and ask: "Use the project memory generated by Knowledge Graph to summarize what was changed previously, what conventions were learned, and what files I should inspect first before continuing."
Additional Comments
Install: clone the repo and run the documented bash installer/setup steps with jq available. Uninstall: remove the installed hooks/scripts and delete the generated knowledge artifacts as described in the README. Network usage: GitHub only for clone/update; runtime behavior is local and git-native. The repo is over 3 weeks old.
Recommendation Checklist
Display Name
Knowledge Graph
Category
Tooling
Sub-Category
General
Primary Link
https://github.com/hilyfux/knowledge-graph
Author Name
hilyfux
Author Link
https://github.com/hilyfux
License
MIT
Other License
No response
Description
Built on Anthropic internal engineering practices and Karpathy's AutoResearch methodology. A zero-dependency, git-native memory layer for Claude Code that persists learned context across sessions. Pure Bash with ~3ms/event overhead, privacy-first.
Validate Claims
Install the tool in a small repo, start a fresh Claude Code session, perform a few edits, then start a second session and ask Claude to recover prior file-level learnings and decisions from the generated project memory. This demonstrates persistent context across session boundaries without a database.
Specific Task(s)
In a test repo, make a few changes across files in session one, then open session two and ask Claude to summarize the previous session's learned conventions and touched files from the generated knowledge artifacts.
Specific Prompt(s)
After installing Knowledge Graph, make a few edits, end the session, start a new Claude Code session, and ask: "Use the project memory generated by Knowledge Graph to summarize what was changed previously, what conventions were learned, and what files I should inspect first before continuing."
Additional Comments
Install: clone the repo and run the documented bash installer/setup steps with jq available. Uninstall: remove the installed hooks/scripts and delete the generated knowledge artifacts as described in the README. Network usage: GitHub only for clone/update; runtime behavior is local and git-native. The repo is over 3 weeks old.
Recommendation Checklist