Chat · Context · Memory
The Ask Vara panel on the right is collapsed by default. Open it from the right-hand rail or with ⌘/ (Ctrl+/ on Windows); on mobile, tap the center button at the bottom. Ask about the current experiment, or have it edit experiments, change parameters, add steps and parse PDFs. Every change shows a confirmation card first, and you decide.
Context selection
- URL auto-infer: On
/experiments/[id]or/materials/[id], the AI uses it as context automatically (gray chip, not removable) - Manual @ pick: Click [@ Pick context] to open the picker with 2 collapsible sections (Experiments / Materials); a pick shows a blue chip with ✕ to remove
Once picked, Vara injects the full data snapshot (params / steps / recipe / recent actionLogs), so no tool calls are wasted.
Per-experiment chat memory
Each experiment remembers its own conversation. Switching to another experiment is a separate conversation; within the same experiment, switching detail ↔ steps does not reset the conversation.
What it reminds you about
Vara is a tool for recording and running experiments; the scientific calls are yours. Ask Vara only brings up execution and record-keeping issues on its own:
- A parameter has no value, or the same parameter uses different units in different places
- Schedules or timings clash, such as two steps booked for the same slot
- Work that was done but has no run logged yet
It won't critique your experimental design or suggest adding controls or replicates unprompted. When you ask about the design or how to read results, it gives the evidence and the options, and you decide.
User memory (ChatGPT-style)
- AI suggests: When it notices a preference in chat, a "Should I remember this?" card appears under the reply. It is saved only when you click Remember; No thanks or ignoring discards it. Nothing inside documents or images is saved on its own
- Manual add: Add them by hand under
Settings → User memory - Cap 50 items / 200 chars each, with LRU eviction
Want to switch models? See Models & Quota.