clef-omni

Usage

const res = await fetch("https://clefdemo.com/v1/clef-omni/run/", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.CLEF_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    model: "clef-omni",
    state: "Checkout has been failing for every customer for the last hour.",
    questions: {
      urgent: {
        type: "noul",
        instructions: "Is this support request urgent?",
      },
      team: {
        type: "choice",
        instructions: "Which team should handle this request?",
        criteria: {
          billing: "Payments, invoices, and refunds",
          technical: "Outages, errors, and configuration",
          sales: "Plans and upgrades",
        },
      },
      severity: {
        type: "score",
        instructions: "How severe is the customer impact?",
        criteria: ["No impact", "Minor", "Major", "Critical"],
      },
    },
  }),
});

const data = await res.json();
// data.answers.urgent / team / severity
console.log(data);

Parameters

• modelstringrequireddefault: "clef-omni"

  Model selector: "clef-omni" (default), also accepts "clef" and "clef-flash". Maps to Workers AI @cf/cloudflare/* on our servers.

• statestring | object | arrayrequired

  Content to evaluate: plain text or structured JSON (records, chat logs, app state). Long text may be truncated to fit context.

• questions {}objectrequiredminProperties: 1, maxProperties: 64

  Map of question id → typed question (noul | choice | score). Ids: letters, digits, _, ., - (max 100). Answers are returned under the same ids.

• images []arraymaxItems: 4

  Optional embedded PNG/JPEG/WebP data URLs (max 4; ≤4 MiB each). Remote URLs are not accepted. Images bill as input tokens.

• audio []arraymaxItems: 4

  Clef-Omni: optional embedded audio data URLs (max 4; ≤8 MiB / 300s each). Combined audio+video decoded ≤16 MiB. Remote URLs not accepted.

• videos []arraymaxItems: 2

  Clef-Omni: optional embedded video data URLs (max 2; ≤16 MiB / 60s each). Remote URLs not accepted.

Authentication

Header: Authorization: Bearer sk_live_…
Create keys in Account. Keys are stored hashed; the plaintext is shown once. Revoked / disabled accounts get 401.

Errors

• 401 unauthorized
• 402 paywall — buy credits on Pricing
• 415 bad Content-Type (must be application/json)
• 429 rate_limit / daily_cap
• 400 validation
• 413 payload too large
• 502 / 504 upstream / timeout — not charged

Security & anti-abuse

• No anonymous access to /v1 — Bearer key required on every call.
• Per-IP and per-key rate limits (≈40 req/min) plus a daily call cap per key.
• API keys hashed at rest; Cloudflare tokens are never exposed to browsers.
• Body size cap (~12 MiB); media must be embedded data URLs (no remote URL fetch).
• Failed upstream inference does not deduct credits.
• Cache hits are free and do not re-run the model.

Billing

1 credit ≈ 1 input token (including media tokenization). Output probabilities are not billed. Packs start at $5 — see Pricing.