Most software forgets you the moment the session ends. BioMate is designed around the opposite premise: that research is cumulative, that every session builds on the last, and that a platform should get measurably better at serving a lab the more it is used.
Why Memory Matters in Research
A bioinformatics platform that treats every session as a blank slate forces researchers to re-establish context every time they open it. Which genome build does this lab use? What QC threshold did we settle on for our cryo-EM particle size? Which ADMET liabilities are dealbreakers for our therapeutic area? This background knowledge lives in researchers' heads, in lab wiki entries, and in the institutional memory of senior postdocs — but not in the software they use every day.
The consequence is repeated re-entry of the same context, inconsistency between sessions and between team members, and a platform that performs exactly the same for a lab on its hundredth analysis as it did on its first. The system does not accumulate any knowledge of what works for this particular team in this particular research context.
Five Tiers, Five Time Horizons
BioMate maintains memory across five distinct levels, each covering a different time horizon. Immediate conversation context captures what has been said and decided in the current session. Session state records the workflows used, parameters set, and results reviewed in a single work period. Project memory holds the key decisions, QC thresholds, and workflow choices that define an ongoing research project — the kind of information that would go into a lab protocol. Lab-level memory accumulates preferences and norms across all projects in a lab or team. Institutional patterns emerge from aggregated signals across similar research contexts, informing recommendations for new users starting similar work.
"After a few sessions, BioMate knows your genome build, your preferred QC thresholds, and the workflows your lab has validated for your data types — without you having to specify them each time."
How Memory Surfaces in Practice
Memory is not a settings page you configure manually. It is inferred from behavior and made available at the point of use. When you start a new RNA-seq analysis, BioMate recalls that your lab works with mouse samples, uses a specific reference annotation, and applies a particular minimum read count threshold — and pre-fills the workflow configuration accordingly. When a parameter differs from what your lab has used before, BioMate notes the deviation and asks whether it is intentional.
Further reading: Lewis et al. 2020, retrieval-augmented generation, Ewels et al. 2020, nf-core Nature Methods, and National Library of Medicine.
The 20th analysis in your lab is faster, better-configured, and more consistent than the first — because the platform has accumulated the context needed to serve you without asking you to re-establish it every time.