Your experiment is often already written down somewhere — in an electronic lab notebook, a project report, a protocol document, or notes from a colleague. You don't have to re-derive the design piece by piece in the chat: paste the description, and the assistant extracts the experiment from it.
How It Works
Open the assistant — from the SDLabs front page or from a draft experiment — and paste your write-up directly into the chat. It can be as informal or as detailed as you have it: a methods section, a bulleted protocol, a paragraph from an ELN entry.
The assistant reads the text and pulls out the building blocks of an experiment:
Variables — quantities you vary, with their ranges and units, and categorical choices with their options
Objectives — what is measured and whether each result should be maximized, minimized, or hit a target
Constraints — stated limits like composition caps or forbidden combinations
Prior knowledge — observed relationships, promising regions, and known effects, drafted into the expert-context summary
As it works, the draft experiment takes shape in the interface in real time, next to the conversation — the same as when you design step by step.
What Happens with Gaps and Ambiguity
Real-world write-ups are rarely complete, and the assistant doesn't pretend otherwise. Anything missing or ambiguous becomes a question, never a silent guess:
A temperature mentioned without a range — it asks for the bounds
A measured quantity with no stated goal — it asks whether to maximize, minimize, or target a value
A vague limit like "keep the cost reasonable" — it asks you to make it concrete before turning it into a constraint
The assistant never invents values that aren't in your text, and it never asserts something about your system that you haven't confirmed. What it extracts is what you wrote — the rest, it asks.
Review Before Launch
Once the extraction is complete, review the configured experiment side by side with the conversation: check the ranges, the objectives, the constraints, and the expert-context summary against your source. Adjust anything in the chat — "the upper bound on baking time should be 75 minutes, not 60" — and the draft updates.
Launching is always your click. The assistant prepares the draft from your description, but taking it live is your decision.
Tips for a Good Starting Description
Any text works, but extraction is fastest when the description includes:
Ranges with units for every quantity you vary
The full list of options for categorical choices
A clear statement of what you measure and what "better" means for each measurement
Limits written as concrete numbers rather than qualitative statements
Don't trim out background knowledge — observed effects and promising regions are exactly what the assistant turns into expert context, which helps the optimizer focus its search from the first round.
Note: Today the assistant works from text you paste into the chat. Reading uploaded documents and searching the scientific literature are planned for a future release.
Related Articles
Meet Your SDLabs AI Assistant — overview and guardrails.
Designing an Experiment with the AI Assistant — the step-by-step setup journey.
The AI Assistant During a Running Experiment — what happens after launch.