AI Decision

Overview

The AI Decision node sends a piece of content (the state) plus a list of typed questions to TypeSafe AI's Jev model and gets back calibrated, structured answers: a chosen option with a probability for every option and a confidence, a position on a rubric you define, or the probability that a yes/no question is true. Jev does not generate text, so there is nothing to parse and no malformed output to handle — and a call typically completes in well under a second.

Use it wherever a workflow needs a judgment about unstructured content: routing inbound messages to the right branch, scoring a draft before it is sent, or deciding whether something needs a human. Keep arithmetic, counting, and date logic in Condition or Code nodes, and use AI Prompt when you need generated or extracted text.

Question Types

• Choice — pick one option from labels you define (up to 255). Returns choice, probabilities per option, and confidence. A Choice question can route the workflow.
• Score — place the state on an ordered rubric of 2–10 described levels. Returns score (may fall between levels), probabilities per level, legend, and confidence.
• Yes / No — returns noul, the probability (0–1) that the answer is yes. No separate confidence value.

Configuration

• TypeSafe AI Credential — required on every plan. The node always runs on your own key and never uses Falcon Builder AI credits.
• State to Evaluate — templatable. Send only what the questions need; unrelated content lowers accuracy.
• State Format — Plain text, or JSON to keep field names and structure visible to the model.
• Questions — each has a name (used in the output), a type, and instructions. Ask narrow, literal questions; one judgment per question.
• Route On — a Choice question whose answer picks the outgoing branch. The node shows one output handle per option label.
• Minimum Confidence — when set, a route answer below this confidence takes the uncertain branch instead, which is the natural place to connect a Human Approval node.
• Model — jev-latest follows each release; pin jev-1.13.0 once you have tuned thresholds.

Output

The output variable (default decision) holds { model, answers, route, usage }. Reference answers by question name:

• {{decision.answers.intent.choice}} and {{decision.answers.intent.confidence}}
• {{decision.answers.urgency.score}}
• {{decision.answers.wants_refund.noul}}
• {{decision.route}} — the branch that was taken

Example: triage inbound support messages

• State: {{trigger.body}}
• intent (Choice, routes): billing — "Charges, invoices, refunds"; technical — "Bugs, outages, integrations"; other
• urgency (Score): "Can wait", "This week", "Today"
• Minimum confidence: 0.6, with the uncertain handle wired to a Human Approval node

Required Credentials