Structured decision outputs
Clef and Clef-flash return bounded, strictly typed answers with probabilities, allowing software to make consistent classification decisions instead of parsing free-form text.
Clef
Clef is Cloudflare’s open-source family of decision models for fast, typed classification in applications and agent workflows. Clef and Clef-flash run on Workers AI, while a related reinforcement-learning platform lets developers fine-tune Clef with their own data.
Decision models for agent workflows
Clef is Cloudflare's family of open-source decision models for fast, programmatic classification. Rather than generating open-ended text, a decision model evaluates supplied state against defined questions and returns typed results with probabilities. Application code can use those results to route work, trigger an escalation, select a team, or defer to a person.
The family includes Clef, positioned as the more capable precision model, and Clef-flash, positioned for latency-critical decisions. Both are hosted on Workers AI and are compatible with the Jev API. Cloudflare also introduced a reinforcement-learning platform for fine-tuning Clef with customer data, while publishing the models on Hugging Face under Apache 2.0 for local experimentation.
Capabilities
Text and image classification
A vision encoder allows Clef to accept images and classify visual content, extending its decision use beyond text-only classification.
Large context window
Clef has a 64k context window, giving applications more input state to classify than the 32k window cited for Jev.
Workers AI hosting
The hosted models run on Workers AI infrastructure, using Cloudflare's edge GPU network to reduce network latency for applications that place decisions in an agent workflow.
Jev API compatibility
Clef is compatible with the Jev API, so existing integrations designed around that interface can experiment with the hosted models with fewer changes.
Reinforcement-learning fine-tuning
Cloudflare has introduced an RL product that lets customers fine-tune Clef with their own data for particular decision tasks.
Practical uses
Support triage
Pass a customer message and questions such as urgency, responsible team, and severity. The resulting classifications can route the ticket to billing, technical support, or sales and escalate high-impact incidents.
Domain and threat classification
Combine Clef with browser-based page retrieval to classify website domains across categories. Cloudflare describes using this workflow for threat intelligence and for identifying potentially malicious or legitimate domains.
Agent routing and escalation
Place a decision model in the hot path of an agent: gather context, classify the situation, then use the result to trigger an action or hand the case to a human when confidence or policy requires it.
Visual content decisions
Use the model's image-capable classification for workflows where visual content is part of the decision context, rather than limiting the classifier to text inputs.
Custom decision models
Fine-tune Clef with organization-specific data when a general decision model does not adequately represent a team's classification task. The source announces this capability but does not describe its training workflow or limits.
Pros and Cons
Pros
- Returns typed classifications and probabilities that can be consumed directly by application logic.
- Supports visual classification through a vision encoder and provides a 64k context window.
- Offers two model variants for different precision and latency priorities.
- Runs on Workers AI and can be used through a Jev-compatible API.
- Open-source model artifacts are available on Hugging Face under the Apache 2.0 license.
Cons
- The source does not provide pricing, training limits, or detailed availability information for the reinforcement-learning platform.
- Benchmark results and latency figures are reported by Cloudflare and cover selected evaluation suites; they should not be treated as a guarantee for every workload.
- The source gives examples of typed question formats but does not document a complete input/output schema or SDK reference.
FAQ
What does Clef do?
Clef is a decision model for classification tasks. It accepts state and questions, then returns typed answers with probabilities that application code can use for routing, escalation, action, or human review.
Where can Clef be used?
The hosted models run on Workers AI. Cloudflare also publishes Clef and Clef-flash on Hugging Face under the Apache 2.0 license for local experimentation and deployment.
What inputs and outputs does Clef support?
Clef supports image inputs through a vision encoder, while the source describes its decision outputs as strictly typed. The example API uses question types including binary-style decisions, choices, and scores.
What is the difference between Clef and Clef-flash?
Clef is the larger model intended for higher-precision decisions, while Clef-flash is positioned for latency-critical decisions. Both are described as compatible with the Jev API.
Can Clef be fine-tuned?
Cloudflare's announced reinforcement-learning product allows customers to fine-tune Clef with their own data. The available source does not specify pricing, training limits, or general availability details.
Quick Facts
- Category
- AI developer tool
- Models
- Clef and Clef-flash
- Hosted platform
- Cloudflare Workers AI
- Local artifacts
- Hugging Face, Apache 2.0 license
- Primary output
- Typed decisions with probabilities
- Fine-tuning
- Reinforcement-learning platform using customer data
Альтернативы Clef
CreateOS Sandbox
CreateOS Sandbox is an isolated compute environment for running code and agent workloads inside Firecracker micro-VMs. It is designed for workflows that need machine-level isolation, private networking between sandboxes, and programmatic control through SDK, CLI, or MCP.
GitBot
GitBot is a local browser-based workspace for creating reusable coding bots powered by Claude Code, Codex, or OpenCode. It helps developers run those agents across repository folders with persistent threads and configurable permissions.
AakarDev AI
AakarDev AI helps teams manage AI provider access, project-level setups, logs, and analytics from one dashboard. It supports BYOK workflows and lists providers including OpenAI, Google Gemini, Anthropic, Groq, Mistral AI, and Perplexity AI.
Trigger.dev chat agent
Trigger.dev chat agent is a durable AI chat backend for developers building stateful conversations that can survive refreshes, crashes, and long-running turns. It connects with the AI SDK `useChat` flow and runs on managed infrastructure with no timeout on a turn.
ByteAsk
ByteAsk is a terminal-first AI coding agent for C and C++ that edits repositories and verifies changes with the real compiler, debugger, sanitizers, and tests before showing a diff. It offers a free tier plus paid plans, with editor connectors and zero-retention handling described in the source.
Codex Plugins
Codex Plugins bundle reusable skills, app integrations, and MCP servers into workflows you can install in the Codex app or use from Codex CLI. They help extend Codex with connected-service tasks, reusable instructions, and shared team workflows.