Clef Demo
Multimodal decision model
Clef Omni is live — try free, no signup
Clef OmniMultimodal decisions for agents — try free online
Clef Omni is a multimodal decision model (30B MoE / ~3B active) that turns a state and typed questions into decisions. It reads text, JSON, images, audio, or video and returns a probability for every allowed option — not a chat LLM.
Don't burn an expensive LLM on a simple route. Hosted Clef Omni API for classify, route, score, and moderation — System One / Jev-compatible schema, prepaid packs from $5.
Free anonymous runs · Prepaid packs from $5
Clef Playground
Default model: Clef Omni · System One schema: state + noul / choice / score · ~156 tokens
Model
30B MoE (3B active) · text, image, audio, video
Media inputs(image · audio · video)
Images
PNG / JPEG / WebP · max 4 · ≤4 MiB · no remote URLs
Audio
WAV / MP3 · max 4 · Omni
Videos
MP4 / WebM · max 2 · ≤60s
Questions schema
Yes / no probability — no criteria list needed.
Output
Probability scores and a short reasoning summary will appear here after you run Clef.
Why developers use this Clef Playground
Try Clef online for agent workflows — multimodal decision model with typed questions, probability outputs, and Workers AI tutorials in one place.
State → typed questions → decisions
The Clef decision model turns a state plus typed questions into decisions, returning a probability for every allowed option — ideal for agent workflow prototypes.
Multimodal inputs
Feed text or JSON state, plus optional images, audio, and video. Test vision-aware decisions online in this playground.
Jev & System One compatible
Keep the same schema. Swap endpoint and model to migrate from Jev, or compare before you commit.
Developer guides (API, local, comparisons)
Start with what the Clef model is, then dive into decision routing, Workers AI tutorials, and example code.
What is Clef Omni?
30B MoE multimodal decision model — text, images, audio, video.
Clef Omni model page
Capabilities, limits, and links to the playground and REST API.
Clef Omni API Python
requests + Bearer sk_live_ example for state + typed questions.
Clef Omni API pricing
Input-token metering, media costs, and prepaid packs from $5.
Clef Omni image input
PNG/JPEG/WebP data URLs, max 4 images, playground + REST.
Clef Omni audio & video
Embedded clips only — limits and multimodal decision examples.
What is the Clef model?
Plain definition of the Clef model — multimodal decisions, not chat — with a playground link.
Clef decision model explained
How state + typed questions return option probabilities for agents.
Clef 27B model
Size, context notes, local VRAM vs trying Clef online.
What is Cloudflare Clef decision model?
Cloudflare-prefixed overview for Workers AI and agent workflows.
Clef vs Jev decision model
Side-by-side comparison for teams choosing between Clef and Jev — API shape, hosting, and when to switch.
How to run Clef locally
27B VRAM notes, Hugging Face weights, and when to use the online playground instead of local GPUs.
Clef-flash vs Clef benchmark
Latency vs quality tradeoffs — when Flash is enough and when full Clef wins.
Clef model example code
Workers AI inference snippets, System One payloads, and agent-loop glue.
Clef for agent routing
Pick the next tool from a closed list with probability thresholds — sharper than generic workflows.
API Doc — Clef Omni REST
sk_live_ auth, curl/JSON examples, errors, and billing for POST /v1/clef-omni/run/.
Playground docs
Workers AI binding sketches and playground payload tips.
Where teams try Clef online
Commercial landing scenarios you can prototype in the playground today.
Clef for agent routing
Score the next tool from a closed list with confidence thresholds.
Ticket classification
Score triage options from ticket state — text or images in the same run.
Security decision gates
Prototype allow / challenge / block style gates with typed questions.
Clef Jev API compatibility (System One)
Clef follows the System One API used by Typesafe’s Jev. Existing integrations can switch by changing the endpoint and model — same state, noul / choice / score questions, and structured probability answers. Read the Clef vs Jev guide.
FAQ
- What is Cloudflare Clef model?
- Cloudflare Clef is a multimodal decision model on Workers AI. It reads a state plus typed questions and returns probabilities for each option — not open-ended chat. See What is the Clef model, or try it in this Clef Playground.
- What is a decision model (Clef)?
- A decision model scores predefined choices given context. The Clef decision model is built for agent routing: state in, probabilities out, Apache 2.0 licensed. Read Clef decision model explained for the full pattern.
- What is Clef-flash?
- Clef-flash is the smaller, faster sibling of full Clef (27B). Use Clef-flash for lower latency prototypes; use full Clef when you want the larger multimodal decision model. See our Clef-flash vs Clef benchmark guide, or compare both in the playground model toggle.
- How do I use Clef for agent routing?
- Give Clef a closed list of next tools or branches as choice options, then threshold the top probability before auto-routing. See the Clef for agent routing guide and Clef model example code for copy-paste payloads.
- Clef vs Jev — which is better?
- They share System One API compatibility, so migration is mostly endpoint + model. Clef runs on Cloudflare Workers AI; Jev is Typesafe’s offering. See our Clef vs Jev decision model guide for a detailed comparison.
- Is Clef compatible with Jev and System One?
- Yes — Clef Jev API compatibility means the same state and noul / choice / score questions. Swap endpoint and model to move an existing Jev integration.
- Can Clef process images?
- Yes. Clef is a multimodal decision model. This playground accepts up to four images with your state so you can test vision-aware decisions online.
- What is the Clef 27B model context window?
- This playground caps estimated input around 8,000 tokens per request for predictable demos — see the Clef 27B model guide for size and local vs online notes.
- Is Cloudflare Clef Apache 2.0?
- Yes. The Clef model is available under the Apache 2.0 license. This site is an independent playground, not affiliated with Cloudflare.
- Do I need GPU to run Clef?
- Not for this playground — try Clef online via Workers AI. To run Clef locally (Hugging Face / 27B VRAM), see our local deployment guide; local GPUs are heavy compared with the online playground.
- How to fine tune Clef model?
- Fine-tuning is an advanced path outside this demo. Most teams start with prompt/state design and the System One schema. Use the playground for inference examples before investing in fine-tunes.
- What tokens are charged?
- Clef only bills input tokens. Output probabilities do not consume tokens. Identical cached runs in this playground are free.