UNDER THE HOOD
Elnora is the AI agent for lab work, built as a harness around the models: a knowledge base of your own science, tools that take real actions, skills that say how to use them, and a sandbox where the code runs.
Elnora is the harness built around it: the memory of your lab, the reach into public science, the place code runs and the methods it follows. Each part does a job a scientist would otherwise do by hand.
MODELS
Elnora runs on frontier and open-weight models and routes each task to the size of brain it needs: the hard reasoning to a large model, the routine turns to a small fast one, which is where most of your token cost hides. It moves between providers mid-task, so a rate limit at one of them slows the work where it used to stop it. Bring your own key, or point Elnora at a model on your own hardware. We are not in the model business, and the less you spend on tokens the better.
KNOWLEDGE BASE
Put your protocols, results and instrument exports in one place and Elnora reads them before it writes anything. Your team works on the same files, every save keeps the version before it, and each report carries the record of how it was made. The more it holds, the better everything it writes gets.
TOOLS
A tool is how the agent takes a real action: run the statistics, fit the curve, decode the plate reader file, search the patent record, resolve a catalogue number. Tools return facts, which keeps an answer current. If the one you need is missing, tell us and it goes in.
SKILLS
A tool is only as good as knowing when to reach for it. A skill is a written method saying which tools a job needs and in what order. Bring your own, or work through something once together and ask Elnora to keep what worked.
SANDBOX
Elnora is a coding agent, and most of what it does for you it does by writing code. The sandbox is where that code runs, with the scientific stack already installed. An agent reasons in probabilities, and this is how it does arithmetic you can rely on, and how it reaches your instruments.
MEMORY
Elnora keeps what it has learned about how you work: your cell lines, your conventions, the shape of a protocol your team accepts. Run the same method three times and it saves that method into your own toolbox as a named skill.
SPECIALISTS
A statistician works out which test your data supports and reports the effect size alongside the p-value. A design-of-experiments planner returns a run sheet in real units. A reviewer grades the work Critical, Major or Minor. Researchers and analysts come in fast or thorough, so routine work goes to a fast one. Several work at once.
INTEGRATIONS
Your scientists open Elnora in a browser and your engineers call it from their own code. The agents your company already runs reach it as a tool over the Model Context Protocol. On the bench it drives connected instruments, holding until a person confirms the deck.
GUARDRAILS
A reference you can follow is one it read, because what reaches the answer came back from a real tool call. A reagent number arrives as a link to the vendor's own product page. Every run can be read back afterwards.
PUBLIC SCIENCE
OPEN DEVELOPMENT
New skills, tools and integrations ship in batches, and the changelog carries every one of them.
Follow the changelogAn AI lab protocol generator turns a described experiment into a plate map, a materials list with resolvable catalogue numbers and a set of deck-specific instructions, while a scientist keeps deciding what the experiment should test.
Carmen Kivisild
Protein structure prediction, image analysis and AI-written research code are already real. A lab that wants to automate lab data analysis with an AI agent needs one that reads the files its own instruments write.
Carmen Kivisild
A scientist's job is a bundle of tasks that are changing at different speeds, and the labs that adopt now, while still teaching the underlying skill, are the ones that keep both the speed and the judgement.
Carmen Kivisild
