A familiar modeling task can involve several small, interruptive steps: find a molecular component, select it, change its representation, inspect a region, then repeat. The challenge is not always the science—it is remembering where each action lives and translating an intention into the right interface steps.
SAMSON AI offers a way to describe some of these tasks in natural language. The key is choosing the command that matches the kind of work you want done: an action in the current document, a selection, or a reusable script.
First, open the assistant
Open SAMSON AI from Interface > Assistant, or use Ctrl+0 on Windows and Linux, or Cmd+0 on Mac. Type / in the input to see the available commands.
Three commands for three different jobs
Use /do for an action. This is suited to requests that change or act on the current document. For example, you might ask it to zoom to the binding site, erase surfaces, or remove water. A request can combine an object and an operation: /do select the ligand and apply a licorice model.
Use /select when the hard part is defining what to select. Instead of manually constructing a selection, describe it in terms of the molecular system: /select all lysines in chain B. SAMSON AI displays the corresponding Node Specification Language (NSL) and executes it. That can be useful both for getting the selection done and for seeing how a natural-language description maps to a formal query.
Use /script when you want Python code. This command generates scripts that use the SAMSON Python API to interact with the document. Examples include translating atoms, creating a small GUI for a repeated operation, or computing the receptor’s gyration radius. Scripts are a better fit when you need a custom or repeatable procedure than when you simply want one immediate interface action.
A practical way to phrase requests
Make the target and the desired result explicit. “Select the ligand and apply a licorice model” is more actionable than “make this look clearer,” because it identifies both what to work on and what to do. For a selection, include useful identifiers or criteria—for example, a chain, residue type, or distance from a ligand. For a script, state the operation and any important units or parameters, such as translating atoms in the z direction by 1 angstrom.
It also helps to work in stages when a task has several parts. First ask for a selection, inspect what is selected, and then request the visual or structural change. This makes it easier to catch an unclear request before it affects a broader part of the document. Treat the result as something to check, not as a substitute for checking: the documentation notes that AI responses can misinterpret instructions or be inaccurate.
When the information is in a document
If you need help based on a web page or PDF, use /learn with its URL, then ask about that material with /refer. For example, you can learn from a paper and then request a summary or a table based on its contents. SAMSON AI can read regular pages and PDFs, and learned information remains available after SAMSON is closed until you use /forget. Use that command when you want to clear all learned information.
Keep in mind that SAMSON AI may not know recent developments and can produce incorrect information. Verify scientific conclusions and critical details against trusted sources. The privacy policy also states that queries and interactions may be collected to improve the service; messages are encrypted during transmission, but absolute security cannot be guaranteed. Consider that when deciding what information to include in a prompt.
The simple distinction is: /do for an action, /select for defining a selection, and /script for generating Python. Choosing the right starting point can make an everyday modeling task easier to express—and easier to review afterward.
Read the SAMSON AI documentation to learn more about commands, limitations, and privacy.
Note: SAMSON and all SAMSON Extensions are free for non-commercial use. Get SAMSON.