chen-006/meow-llm-detector

用于检测ai模型是否真实

What it solves

It identifies the underlying behavior of Large Language Models (LLMs) to help users determine if a model is actually the one claimed by a provider, particularly when using API intermediaries or proxies. It provides "fingerprint evidence" rather than absolute identity certification.

How it works

The tool sends a series of short questions to a target model and analyzes the distribution of its responses. By comparing these response patterns against established benchmarks, it can estimate the match between the target model's behavior and known model families or specific versions.

Who it’s for

Developers and API users who want to verify the authenticity of the models they are accessing via third-party providers or API gateways.

Highlights

  • Multi-platform support: Available as a portable Windows version or source code for macOS and Linux.
  • Automated Benchmarking: Automatically checks for and updates benchmarks from the maintainer every 24 hours.
  • Flexible Configuration: Supports custom API URLs and keys, and allows users to select from predefined model lists or enter custom request names.
  • Privacy-focused: Temporary keys are kept in memory, and saved credentials are stored in the system credential store.

Related

  • Project
    hanlinwenyuan/hlwy-ai-checker

    A locally‑run Python tool that statistically fingerprints an official LLM API and compares it with a third‑party API to detect possible tampering, using repeated prompts and low token consumption.

  • Project
    xqy2006/ModelTrace

    ModelTrace is a Python web tool that identifies the specific LLM (GPT or Claude) behind a response by sending three integer‑generation challenges, extracting numeric fingerprints, and matching them against a local library of known model fingerprints. It supports manual paste‑in testing and automated API calls, includes a Codex plugin for background monitoring, and ships with a pre‑built bank of 18 model versions.

  • Project
    EvolvingLMMs-Lab/lmms-eval

    A unified evaluation toolkit for Large Multimodal Models that provides reproducible and efficient benchmarking across image, video, and audio tasks.

  • Project
    Yinsongxu/LLM2Jev

    A framework that converts local LLMs and VLMs into structured decision models by extracting probabilities from logits during prefill, bypassing token-by-token decoding for lower latency.

  • Project
    llm-as-a-verifier/llm-as-a-verifier

    A general-purpose verification framework that uses fine-grained logprob-based scoring to evaluate and rank AI agent trajectories across text and multimodal tasks.