What to put in memory
Memory works best for short, stable facts that change how an analysis should be run — not for transient state from a single session. Good memory entries:- “We always export supplementary tables as TSV, not CSV.”
- “For mouse models, use GRCm39 unless the dataset specifies otherwise.”
- “Our strain panel is BL6, BALB/c, and 129S1 — always include all three in PCA plots.”
- “For DESeq2, default to a 10% FDR threshold for our preliminary reports.”
- “Cite the 2024 lab style guide when generating methods sections.”
- File contents or sample-level data.
- Anything specific to a single experiment.
- Anything that changes week to week.
Adding a memory entry
The simplest way to add an entry is to tell the agent in plain English:1
Open the workspace Memory tab
Click Memory in the workspace sidebar.
2
Search or browse
Existing entries appear as a searchable list. Each entry shows the fact, the kind (manual note, derived from a conversation, derived from a tool output), and the date it was created.
3
Edit or remove
Click any entry to view its detail. From here you can rewrite it inline or delete it. There is no version history — edits take effect immediately.
How the agent uses memory
Before starting a run, the agent reads the relevant memory entries and treats them as binding constraints — the same way it would treat an attached Skill. A memory entry that says “always use TSV” makes the agent default to TSV for any supplementary file it produces, even if a teammate’s prompt is silent on the format. Memory is consulted, not enforced. If a prompt is explicit (“export this one as CSV”), the prompt wins. The memory entry still applies to the next request.Quota
Memory is capped at 5,000 entries per workspace. The cap is generous for normal team usage — entries are sentences, not documents — and is visible as a quota bar in the Memory tab. If you approach the limit, look for redundant or outdated entries and remove them.A few tips
- Promote, don’t repeat. If you find yourself typing the same caveat into three different prompts, move it to memory.
- Review periodically. Memory entries that were true six months ago may not be true now. A 10-minute monthly skim keeps the store tidy.
- Be specific about scope. “For mouse RNA-seq, use GRCm39” is more useful than “Use GRCm39” — the agent applies entries to the situations they describe.
- Memory is shared. Anything you write is visible to every member of the workspace. Don’t put anything in memory that you wouldn’t say in a lab meeting.
Related pages
Workspaces
The nine workspace tabs and where Memory fits among them.
Skills
Reusable building blocks the agent applies during a run.

