Use the WEEMS Agent

The WEEMS Agent can explain project work, create or update notebooks and files, run calculations, and review results.

What can I ask the WEEMS Agent to do?

Ask it to explain, create, change, run, or review work in the project you have open.

Common requests include:

• explain a notebook, file, chart, error, or result;
• create or organize a notebook;
• write, correct, or explain code;
• analyze data and show the important findings;
• prepare calculations, simulation steps, or a report;
• make a small, named change to an existing file; or
• review work and list possible mistakes without changing anything.

If a task needs a running machine, WEEMS also needs either a cloud machine or Local Compute. Running work can use compute separately from the agent conversation.

How do I start my first conversation?

Example:

Review the notebook named Cooling Loop. Do not change anything yet. Explain why the pressure result jumps after row 20, use SI units, list your assumptions, and recommend one next step.

You will see an answer in the chat and, when relevant, a short record of files read, work run, or changes proposed.

What makes a good request?

A useful request says what you want, where to work, what rules to follow, and how you will know it is finished.

Use this quick pattern:

1. Outcome: “Create a one-page summary of the test results.”
2. Location: “Use the notebook named Pump Test and the file results.csv.”
3. Rules: “Use SI units. Keep the raw data unchanged. State every assumption.”
4. Boundary: “Change only the Summary section.”
5. Finish line: “Include a table of the three largest differences and a short conclusion.”

Put it together like this:

Using @Pump Test and @results.csv, update only the Summary section. Keep the raw data unchanged, use SI units, and state every assumption. Add a table of the three largest differences and finish with a short conclusion. Check the numbers before you stop.

Type @ in the message box and choose a notebook or project file from the list when the name could be unclear. You can also open the + menu and choose Attach files or Attach images for material that is not already in the project.

If you only want advice, say “Do not change or run anything.” If the task is large, first ask for a plan, then approve one small stage at a time.

How do I use the WEEMS Agent from a cell?

The cell-level Agent action—also called Cell AI—adds one cell to an Agent conversation. In the current interface, the sparkle button is labeled Edit with AI.

Clicking the cell toolbar button does not send a request or change the cell by itself. It adds the cell content to the message box so you can give the Agent an instruction.

How do I edit highlighted text or code?

Highlight to edit makes a proposed replacement inside one code or text cell. The original selection remains unchanged until you accept the proposal.

Highlight to edit applies only to the selected range. Use the cell toolbar workflow when the request needs the whole cell or a longer Agent conversation.

What do “Ask before edits” and “Proceed automatically” mean?

The control label below the message box determines whether WEEMS pauses before the agent acts.

When WEEMS pauses, you may see two different cards:

• Plan ready describes the overall approach. Choose Approve to continue, or Reject to send feedback. Approving a plan does not mean every later action is automatically safe.
• Approval Required describes a specific action and the affected items. Choose Approve only when the action, target, and reason match your request. Choose Deny to stop that action.

Changing the control label affects requests sent with that selection. Check it again when you start another chat or return later.

What should I check after the agent finishes?

Use this one-minute check before you rely on the result:

1. Open every file or notebook the agent says it changed.
2. Confirm it changed only the items you named.
3. Check the input data, units, assumptions, signs, and decimal places.
4. Re-run an important calculation and confirm it completes without an error.
5. Compare safety-critical or regulated work with the approved method, standard, or a qualified reviewer.

Use this follow-up request to produce a review summary:

Summarize exactly what you changed, what you did not check, every assumption you made, and the three things I should verify myself.

If a change is unexpected, do not continue building on it. Tell the agent what is wrong, or use the project's undo or history controls before making more changes.

How do I stop or correct the agent while it is working?

• Correct it: Type the new instruction in the message box and send it. The agent uses the message to redirect the work in progress.
• Stop it: When the message box is empty, click the square Stop button where the send arrow normally appears.
• Answer it: If the chat says Plan ready or Approval Required, the agent is waiting for your choice. Approve, reject, or deny the card to continue.
• Narrow the task: If progress is slow or unfocused, send a smaller instruction such as, “Stop after reviewing the first notebook. Do not make changes.”

Stopping prevents more work from starting, but it does not remove changes that already finished. Review the project before sending another request.

How do I choose a model, files, and web search?

The controls below the message box determine the model and information the agent can use:

• Model: Click the model name to see the available choices. Models vary in speed, cost, and capability. If the request does not require a specific model, keep the selected option.
• Project file or notebook: Type @, start typing its name, and select it from the list. This is more reliable than referring to “that file.”
• File or image from your device: Click +, then Attach files or Attach images. Wait until the attachment is ready before sending.
• Web search: Click + and switch Web search off when you want the answer based only on your project. Web search is on when a new composer opens.

Models currently provided by WEEMS

WEEMS currently provides these nine models. Your plan controls how much WEEMS-hosted use is included.

The model picker is the final list for your account. If a model in this table does not appear there, refresh the page and check your plan or saved provider key.

The model list can differ by account and by the provider keys saved in the browser. To add your own provider key, open Security from account settings, find Provider API keys, add the matching key, and click Save. The saved key stays in that browser, so clearing browser data or using another browser removes it from WEEMS.

When you use your own provider key, that provider charges your account for the model request. WEEMS compute and storage charges can still apply if the task runs a machine or stores new material.

Usage, privacy, and limits

• Your plan controls how much WEEMS-hosted agent use is included. Longer conversations, larger material, and some model choices can use more of that allowance.
• The chat can show the latest usage and cost near the message box after a response. Open Billing from account settings to review your plan, compute use, storage, and prepaid balance.
• Running a notebook or calculation can use compute in addition to agent usage. See Local Compute to run the calculation on your own computer.
• The agent may use the message, attachments, and relevant content from the current project to complete the request. WEEMS limits it to project content that your account is allowed to open.
• Do not paste passwords, provider keys, private access details, or personal information that the task does not need.
• When Web search is on, the agent may use information from outside the project. Ask it to show sources and dates, then check them before relying on the answer.

Provider terms may differ, especially when you use your own provider key. Follow your organization's rules for confidential or regulated information.

Troubleshooting

Best practices

1. Start with Ask before edits.
2. Name the project item with @ instead of saying “this” or “that.”
3. Include units, limits, and anything that must remain unchanged.
4. Ask for one clear result at a time.
5. Review changed work before approving the next step.

FAQ

Can I ask for an explanation without changing anything?

Yes. Say, “Explain this without changing or running anything,” and use Ask before edits.

Can I return to an earlier conversation?

Yes. Open Chats in the project and select the earlier chat from the list. Click New chat when you want a separate conversation.

What should I send support if the problem continues?

Include the project name, chat title, selected model, approximate time, exact error message, and a screenshot of the final status. Never include a provider key, password, or other secret.