Memory
Version: 2.5 | Last updated: 2026-10-26
Open Memory from the sidebar to see what Bluebox has learned from your investigations.
Learnings stay scoped to your workspace. Bluebox never uses them to train a shared or custom model, and never applies them to another workspace's investigations.
Review learnings
After each completed investigation, Bluebox distills learnings—patterns, preferences, and facts it noticed that are likely to help future investigations. Each learning shows:
- Learning—the rule Bluebox distilled, in plain language.
- Influence—a 0–10 score for how strongly Bluebox weighs the learning, based on its importance, how often other investigations found it again, and how it has been rated. The score changes only when the learning's own evidence changes. A learning Bluebox has not used or rated yet shows Not used yet instead of a score.
- Usage—how many times the learning was shown to Bluebox (shown) and how many times Bluebox actually used it (applied).
- Feedback—how the learning has performed: Helpful, Mixed, Misleading, or No feedback yet.
- State—In use or Muted, plus a marker when a workspace owner has manually kept or muted it.
Use the All / In use / No feedback / Harmful / Muted filter to narrow the list, search by keyword, filter by topic, and sort by influence, most applied, or most harmful. No feedback and Harmful cover only learnings still in use—muting a learning moves it to Muted immediately, whatever the reason, and removes it from both.
The list loads up to 200 learnings; if your workspace has more, a note above the list tells you how many are not shown.
Select a learning to open its detail page. It repeats these signals, lists the parts that add up to its influence score, and shows the full rule Bluebox distilled—what to do, what to avoid, and any additional detail. When harmful ratings outweigh everything else, the score stays at 0.
Mute or unmute a learning
Only workspace owners can mute or unmute, from the learning's row menu or its detail page.
To mute a learning, select Mute, choose a reason—Outdated, Wrong or misleading, Too narrow, or Noisy, low value—optionally add a note, and confirm. A note is visible to every workspace member, so don't include credentials, tokens, or other secrets in it. Muting excludes the learning from every future investigation and pins it so Bluebox's own auto-mute step leaves it alone from then on.
To bring a learning back, select Unmute and confirm. Unmuting pins the learning too, so it won't be muted again automatically afterward.
Next steps
- Investigations—learnings apply automatically during investigations.
- Workspaces—workspace owners control which learnings stay active.