Clef model use cases.

01 / ticket-routing

Support ticket routing

Choose a queue, rate impact, and flag review.

Clear example

Our team cannot sign in after the latest update. All twelve users are blocked.

Boundary example

I cannot sign in, but I may have used my old email. Also my invoice seems different.

Questions to ask

{
  "queue": {
    "type": "choice",
    "instructions": "Which support queue should handle this ticket?",
    "criteria": {
      "technical": "Login, errors, or service failures",
      "billing": "Invoices, payments, or subscription charges",
      "general": "Other questions"
    }
  },
  "impact": {
    "type": "score",
    "instructions": "Rate the reported business impact.",
    "criteria": [
      "Minor inconvenience",
      "One person blocked",
      "Team or critical work blocked"
    ]
  },
  "review": {
    "type": "noul",
    "instructions": "Does this ticket require a human support agent?",
    "criteria": {
      "true": "Blocked work or unclear issue needs a person",
      "false": "A routine answer is sufficient"
    }
  }
}

What happens next

Send to the chosen queue; prioritize higher impact and human review.

Limit: Mixed issues and incomplete reports can be misrouted. A person should review uncertain cases.

The preset result is illustrative, not a measured model response. Test both examples against your own review rules before routing real work.

02 / agent-routing

Agent next step

Choose among explicitly allowed actions.

Clear example

The user asks for the current delivery status of order A104. The order lookup tool is available.

Boundary example

Please cancel the thing I ordered last month, or just tell me where it is.

Questions to ask

{
  "next_step": {
    "type": "choice",
    "instructions": "Choose the next allowed action.",
    "criteria": {
      "lookup_order": "Look up an existing order status",
      "answer_faq": "Answer a general policy question",
      "human_review": "Escalate unclear or unsupported requests"
    }
  }
}

What happens next

Only execute allowed actions; apply a separate confidence threshold and human escalation rule.

Limit: A chosen label is not authorization. Validate tool arguments and user permissions separately.

The preset result is illustrative, not a measured model response. Test both examples against your own review rules before routing real work.

03 / content-triage

Content review

Check whether a post needs human review, using sample rules.

Clear example

Here is my step-by-step guide to growing tomatoes in balcony containers.

Boundary example

That competitor is killing us. Someone should teach them a lesson.

Questions to ask

{
  "category": {
    "type": "choice",
    "instructions": "Classify under the example rules. These are illustrative rules, not a platform policy.",
    "criteria": {
      "routine": "Ordinary informational content",
      "review": "Ambiguous claims or context needing human review",
      "restricted": "Explicit threats of violence against a person"
    }
  }
}

What happens next

Queue ambiguous or restricted content for human review before any enforcement.

Limit: Context, quotation, and sarcasm can change meaning. This initial screen can produce false positives and false negatives.

The preset result is illustrative, not a measured model response. Test both examples against your own review rules before routing real work.

04 / image-checklist

Product image checklist

Check clarity, complete framing, and a plain background.

Clear example

An original illustration of a blue bottle on a plain light background. Check the example listing requirements.

Boundary example

The bottle may be cropped at the top; reflections make the outline hard to judge.

Questions to ask

{
  "clear": {
    "type": "noul",
    "instructions": "Is the product clearly visible?"
  },
  "complete": {
    "type": "noul",
    "instructions": "Is the entire product inside the frame?"
  },
  "background": {
    "type": "noul",
    "instructions": "Is the background plain and light?"
  }
}

What happens next

Use the checklist to suggest a manual image review or a new photo.

Limit: This simple illustration does not represent real photography conditions. Borderline framing and reflections need human review.

The preset result is illustrative, not a measured model response. Test both examples against your own review rules before routing real work.

Send a local image from your server

Set IMAGE_PATH and IMAGE_MIME to your local PNG, JPEG, or WebP. The cURL version requires cURL, jq, and the base64 command. The generated request reads and base64-encodes that file; exported playground configurations omit image bytes.

# Set IMAGE_PATH to a local PNG, JPEG, or WebP and IMAGE_MIME to its MIME type.
: "${IMAGE_PATH:?Set IMAGE_PATH}" "${IMAGE_MIME:?Set IMAGE_MIME}"
PAYLOAD='{"model":"clef-flash","state":"An original illustration of a blue bottle on a plain light background. Check the example listing requirements.","questions":{"clear":{"type":"noul","instructions":"Is the product clearly visible?"},"complete":{"type":"noul","instructions":"Is the entire product inside the frame?"},"background":{"type":"noul","instructions":"Is the background plain and light?"}}}'
IMAGE_BASE64=$(base64 < "$IMAGE_PATH" | tr -d '\n')
printf '%s' "$PAYLOAD" | jq --arg image "data:$IMAGE_MIME;base64,$IMAGE_BASE64" '. + {images: [$image]}' | curl "https://api.cloudflare.com/client/v4/accounts/${CLOUDFLARE_ACCOUNT_ID}/ai/run/@cf/cloudflare/clef-flash" \
  -H "Authorization: Bearer $CLOUDFLARE_API_TOKEN" \
  -H 'Content-Type: application/json' --data-binary @-