mrmps/classifier-dev

Zero-shot text classification over plain HTTP — no API key, no account. One Cloudflare Worker, a CLI, and an MCP server. https://classifier.dev

What it solves

classifier.dev provides a simple, zero-shot text classification service that allows users to categorize text without needing to train a custom model or manage API keys. It solves the problem of quickly turning plain text into a specific label and a calibrated confidence score.

How it works

The service is built as a single Cloudflare Worker. It primarily uses TypeSafe's Jev, a decision model (rather than a traditional LLM) that returns calibrated probabilities for given options in approximately 150ms.

For higher accuracy, the "smart" tier re-evaluates answers with low confidence (below 0.7) using a chain of fast reasoning models (such as Gemini-3.8-flash). If the primary decision model is unavailable, the system falls back to a chain of LLMs. It also supports multi-label classification in a single pass and can handle long-context documents (up to 10 million tokens) by chunking and screening for relevant evidence.

Who it’s for

Developers and agents who need a fast, no-signup classification API that can handle both small batches of text and very large documents.

Highlights

  • Zero-Shot Capability: No training or signup required; just provide text and labels.
  • High Throughput: Capable of processing up to a thousand texts per request.
  • Calibrated Confidence: Provides confidence scores that correlate with actual accuracy.
  • Long Context Support: Handles documents up to 100MB using a recursive chunking and screening process.
  • Agent-Friendly: Implements the feedback.now protocol for structured reporting from AI agents.

Related

  • Dispatch
    Jev Model Review: Fast, Cheap Text Classification Across Diverse Tasks

    Jev, the newly released AI model, delivers high‑accuracy, low‑cost text classification comparable to fine‑tuned BERT models while running faster and cheaper than larger LLMs.

  • Project
    razorback16/openjev

    OpenJev is an open-source decision server that provides fast, structured, and calibrated answers (yes/no, choice, score) from diffusion-based models by reading their internal probability distributions.

  • Project
    firelex/jeff

    Jeff is a 0.8 B “System 1” model that makes ultra‑fast, calibrated classification decisions. It uses lightweight LoRA adapters for specific domains (guard, triage, spam, etc.) and can act as a fast‑path filter in front of a large LLM (e.g., Qwen 3.8‑27B), handling most queries in ~30 ms and delegating only uncertain cases. Weights and adapters are open on Hugging Face, GGUF files enable llama.cpp use, and a simple HTTP/Python API (`jeff‑serve`) lets you plug Jeff into any service.

  • Project
    Knowledgator/GLiClass

    A lightweight, zero-shot sequence classification model that performs classification in a single forward pass to be 10x faster than traditional cross-encoders.

  • Project
    fnnx-ai/scikit-llm

    A library that integrates Large Language Models like ChatGPT into scikit-learn, enabling enhanced text analysis tasks using a familiar ML pipeline API.