SDLabs includes an AI assistant, built directly into the platform, that works alongside you as you design and run optimization experiments. You describe what you want in plain language — the assistant translates it into a correctly configured experiment and helps you make sense of the results as they come in.
Where to Find It
From the SDLabs front page — start a conversation before any experiment exists, and the assistant will build one with you from scratch.
From any experiment — look for the assistant icon in the top right of any draft or running experiment. The assistant can see the experiment you have open, so your questions are answered in context.
What It Helps With
The assistant adapts to where you are in your workflow:
Designing a new experiment. Start from a single sentence and the assistant guides you through parameters, objectives, constraints, and expert knowledge, step by step. See Designing an Experiment with the AI Assistant.
Starting from an existing write-up. If your experiment is already described somewhere — a protocol, a report, an electronic lab notebook entry — paste it into the chat and the assistant extracts the design from it. See Starting from an Existing Experiment Description.
Working with a running experiment. Once your experiment is live, the assistant explains what the optimizer is recommending and why, keeps an eye on model freshness, and refines the guidance with you as your understanding evolves. See The AI Assistant During a Running Experiment.
What Stays in Your Hands
The assistant is designed to be a safe collaborator. A few boundaries are deliberate, and they hold in every conversation:
You launch, the assistant prepares. It never starts, publishes, restarts, or retrains an experiment — those are always your explicit click.
Your data is yours to enter. The assistant never submits measurement values on your behalf, and it never invents them: it will not fabricate measurements, descriptors, or domain claims it cannot ground in what you have shared.
Running experiments keep their shape. The assistant does not change the structure of an experiment that is already running — its parameters, constraints, and algorithm remain under your control in the configuration view. If you want to make such a change, the assistant will guide you to where and how to make it yourself.
When in doubt, it asks. Ambiguous or missing information triggers a clarifying question, never a silent guess.
At a Glance: Can and Can't
The assistant can:
Build a draft experiment end to end — parameters (numerical and categorical), measurements and objectives, batch size, constraints, and the strategy for balancing multiple goals
Work from a pasted description of your experiment — a protocol, report, or notebook entry — and turn it into a configured draft
Add descriptors to categorical options — using the values you provide
Draft and refine the expert context, and improve the experiment's name and description, including on a running experiment
Read the experiment you have open — its configuration and status — to answer your questions about it
Fetch and explain the optimizer's recommended next conditions, and tell you whether the model is up to date or would benefit from a refresh
The assistant can't:
Start, publish, restart, or retrain an experiment — those are your click
Enter, change, or invent measurement values or descriptor values it didn't get from you
Change the parameters, constraints, or algorithm of a running experiment — it guides you to the configuration view instead
See your charts and plots — it reasons from the experiment's data, not the visualizations
Search the scientific literature — planned for a future release
How We Make Sure You Can Trust It
Before every release the assistant is run through an evaluation suite of realistic, multi-turn scientific scenarios. Every run is scored by an independent evaluator against explicit success criteria, on the real assistant and the real platform, not mocks. The behaviors described in these articles are tested before they reach you.
Where to Go Next
Designing an Experiment with the AI Assistant — the guided setup journey, from a single sentence to a launch-ready draft.
Starting from an Existing Experiment Description — paste a write-up from another source and let the assistant extract the design.
The AI Assistant During a Running Experiment — interpretation, model freshness, and refining the guidance mid-run.