How Memory Works
When the assistant spots something worth preserving — a key result, a decision, a useful pattern — it suggests saving it. You see a card with what it wants to remember and an Apply button. Nothing is saved without your approval.
Memory is a shortcut, not a gate. The assistant can always search your notebooks for past results. Memory just lets it recall key facts instantly without re-reading everything.
What gets remembered
Section titled “What gets remembered”- Findings — “Western blot showed 3.2-fold increase in phospho-ERK after 10 min EGF stimulation”
- Decisions — “Switched from RPKM to TPM because cross-sample comparison is more reliable”
- Patterns — “HEK293 transfections below 60% efficiency correlate with passage numbers above 30”
- Preferences — “Use seaborn with the ‘colorblind’ palette for all figures”
- Protocols — “Standard cell lysis: aspirate media, PBS wash x2, RIPA + protease inhibitors on ice 15 min”
Each memory stands on its own — it includes specific values and context so the assistant can use it without needing to look up the original source. A memory like “the experiment worked” is not useful; a memory like “10 nM doxycycline for 24 hours gave 95% knockdown of TP53 in HCT116 cells” is.
Memory is scoped to your project
Section titled “Memory is scoped to your project”Memories are stored at the project level. The assistant in one project cannot see memories from another project. This keeps context clean and relevant — your proteomics project does not interfere with your genomics project.