How to effectivally use it on antigravity or other IDEs?

[ynaponte]

I'm struggling to use it effectivaly and all tutorial only teach how to setup not how to effectivally use it.
First thing i think I did wrong was I've ran /graphify ., but in the chat, meaning I've had to spend tokens for graph creation.
Also, in other requisitions, for example for test plan implementation, with the graph created, the agent did not read the graph, instead choosing to read the entiry codebase again. I don't know what I should've done for it to read the graph.

Summaryzing, what are your tips for usage?

[TPAteeq]

You were right about the first one. Build the graph from your terminal, not chat:

graphify extract . --code-only

Local AST parsing, no API key, nothing in your chat context.

For the agent ignoring the graph: a graph.json on disk doesn't change agent behavior by itself. Something has to tell it to look. That's what the per-IDE installers do:

graphify antigravity install # or: cursor / codex / gemini / kiro / opencode install

For Antigravity that writes .agents/rules/graphify.md as an always-on rule: when graphify-out/graph.json exists, run graphify query "" before grepping or reading files. Other IDEs get the equivalent rules file or hook, see graphify --help. The installer also prints an MCP config block worth adding, since real tools beat hoping the agent shells out.

Then set up freshness once:

graphify hook install

Post-commit and post-checkout git hooks, so the graph rebuilds itself as you work. Works regardless of IDE. (graphify update . is the manual version if you want it current mid-session.)

One thing that helps in practice: the rule is a nudge, not a guarantee. For a task like your test plan, the agent thinks it's implementing, not answering a codebase question. Just say "use graphify to find the affected modules first."

Happy to help :)