Create & connect
Decision nodes: get a probability-based choice
Ask a decision model to compare options, answer yes or no, or rate a scale using a workspace branch as context.
Before you start
Section titled “Before you start”A decision node asks a decision model a closed question about the current branch. Unlike a language-model answer, it returns a selected option and probabilities rather than an open-ended explanation. Decision nodes are available in the app; the feature can be unavailable temporarily, so use the visible Decide control when it appears.
The decision model uses the branch above the decision and can also read documents you attach. It is most useful when the choices are clear and you can describe what “best” means in the question.
When to use this
Section titled “When to use this”Use a decision node to choose among a short list, answer a yes/no question, or rate a choice on an ordered scale. It can provide a quick signal, not a substitute for checking important facts, considering missing options, or making a high-stakes decision yourself.
- In Canvas, hover over an answered node or input node and choose Decide near its +. Alternatively, open ⋮ and choose Decision node. In Chat, choose Decide below an answer or note.
- Write a specific question in the decision form. Mention the criterion you care about, such as “Which option best meets the stated budget and accessibility requirements?”
- Choose Options for a choice among alternatives, Yes / No for a binary decision, or Scale for ordered levels. For Options, enter between two and ten options, one per line. For Scale, enter levels from lowest to highest, one per line.
- If the answer immediately above already lists alternatives, use the suggested-options button if it appears, then check that the options are complete and distinct.
- Optionally attach documents with the paperclip. The node already uses its branch as context; attachments can provide additional evidence in text form.
- Select Decide or use the form shortcut (Ctrl+Enter on Windows/Linux or ⌘+Enter on macOS). Wait for the result: it shows the selected answer, a probability bar for each option, and an overall confidence value.
- Read the probabilities, then continue the conversation from the decision node if you want a language model to explain implications or turn the choice into an action plan.
Worked example
Section titled “Worked example”After a branch describes a small team’s requirements, ask: “Which tool best fits a team of four that needs shared tasks, email notifications, and a free starting option?” Set the type to Options and enter “Tool A”, “Tool B”, and “Tool C”, one per line. Submit, then compare the winning option and the bars. If confidence is low, treat that as a sign the options are close or the context is not decisive; check the requirements and gather more evidence before acting.
What happens next
Section titled “What happens next”The decision result remains a node in Canvas or a turn in Chat. It becomes part of the context for subsequent nodes on that branch. You can branch from it to ask for explanation, add documents or notes, and make another decision with a refined list. If the question is better answered with prose, use a regular model node instead.
Cost and limitations
Section titled “Cost and limitations”A decision node uses a small number of credits, shown on the resulting node. The model classifies among the options you provide; it does not independently verify every real-world claim or guarantee that your list covers all possibilities. The probability values express the model’s distribution across the supplied choices, not objective certainty. Low confidence means you should take extra care. Provide clear options and enough relevant context, and verify consequential decisions independently.
Troubleshooting
Section titled “Troubleshooting”- Decide is not visible: use a saved answer or input node and check again. Feature availability can change; if it remains unavailable, retry later.
- The form will not submit: enter a question and select a model if offered. For Options or Scale, provide 2–10 non-empty lines; Yes / No needs no option list.
- “The decision model couldn’t answer”: wait briefly and try again. If it continues, simplify the question or use a regular model answer.
- The result seems unrelated: check that you started from the intended branch and attached the right document. The decision model reads that branch, not every workspace branch.
- Probabilities are close: the answer is uncertain between the supplied options. Refine the criteria or collect more evidence rather than treating a narrow lead as a guarantee.