Decision AI

Decision AI nodes are fast, focused evaluations that classify, score, or determine outcomes using criteria you define. Their outputs are always one of the values you declared, so the rest of the workflow can route on them. They're built for lightweight, high-volume evaluations and charged a fixed price per evaluation. To write, summarize, or extract content instead, use Generative AI; see AI Nodes for how to choose.

Prebuilt Decision AI

Under AI Nodes → Decision AI in the Studio node library, choose Browse to add prebuilt nodes to a workflow, such as returns, routing, or lead qualification. Each one takes typed inputs and returns typed decisions.

Custom Decision AI

You define typed inputs, optional reference context (your policy or rubric), and up to 8 decisions, right inside a workflow in the Studio: under AI Nodes → Decision AI in the node library, choose Browse, open the Custom tab, and choose Create Decision AI. Nothing is stored until you choose Create node; the node then appears in the library, ready to drag onto the canvas. Select it and choose Edit node to change, test, or publish it later. Your Custom Decision AI nodes appear in every workflow's library; prebuilt ones are added per workflow from the catalog. Each decision is one of:

• Yes / no: outputs yes, no, or uncertain.
• Choose one: outputs one of your options (up to 12) or uncertain.
• Score on a scale: outputs one of your levels (up to 7) or uncertain.

You set the confidence each decision needs. Below it, the output is uncertain, so your code can route those cases to a person. Kovca doesn't return probabilities.

Deploy it as a workflow

Nodes are components; workflows are what you call. A workflow with just Input → your node → Output accepts exactly the node's input and returns exactly its output, so after you publish the node and the workflow you call it like this:

curl https://kovca.io/api/v1/workflows/wf_.../execute \
  -H "Authorization: Bearer $KOVCA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "input": { "message": "Can I return a used blender?" } }'

{
  "request_id": "req_...",
  "run_id": "run_...",
  "workflow": { "id": "wf_...", "version": 1 },
  "status": "succeeded",
  "output": { "eligible": "no", "reason": "used_item" },
  "steps": [ ... ],
  "evaluations": 1
}

Every output field is one of its declared values. Published nodes also appear under Add in every workflow, so you can reuse them in larger workflows. See the Workflows API.

Versions

• Test drafts from the node editor in the Studio (billed like any evaluation). Published node versions are immutable.
• A published workflow must use a published node version, and stays pinned to it. Publishing a new node version doesn't change deployed workflows until you update and republish them.
• Archiving a node removes it from the Add list; workflows already using its versions keep working.

Billing and limits

• Each successful run of the node is one evaluation, however many decisions it makes.
• Node inputs up to 16 KB. A node times out after 15 seconds.