Guides
Tune your memory
Keep personalization understandable.
OpenStar Memory improves when users correct it.
Treat Memory like a living preference file
The goal is not perfect inference. The goal is a profile the user can understand and steer.
Save
Relevant, useful later, but not urgent.
Watch
Important enough to monitor for future changes.
Hide
Off-topic, repeated, or no longer useful.
Ask
Clarify why a card appeared or what to compare.
There should be no fake fallback memory. If OpenStar has too few signals, the app should explain how to build memory rather than pretending it already knows the user.
Tuning loop
Use Explore normally and act on repos instead of only scrolling.
Save and Watch the repos that represent real interest.
Hide noisy clusters so the feed can rerank away from them.
Review Memory when the feed starts drifting or repeating itself.
Signal meaning
| Action | Memory impact |
|---|---|
| Save | Adds moderate positive interest without requiring future briefs. |
| Watch | Adds strong interest and enables retention through briefs. |
| Hide | Adds negative feedback for reranking. |
| Ask follow-up | Adds intent context that raw GitHub stars cannot capture. |
| Edit interest | Overrides weak inferred signals. |