Sniffy, a new free AI detector and Chrome extension, launched on Hacker News today, aiming to provide a zero-cost alternative to paid services like Pangram. The tool scans text to determine if it was written by a human or generated by models such as ChatGPT, Claude, or Gemini. It offers both a web-based checker and a browser extension that overlays detection badges directly onto social media posts.
Under the Hood
The developer built Sniffy by directing Fable 5.2 to fine-tune a RoBERTa model. This approach allows the tool to score text word-by-word rather than analyzing the entire document as a single block. This granular scoring enables Sniffy to identify specific sentences within a mostly human post that may have been polished or generated by AI. The current model is trained on data from human writing and outputs from major language models, including Llama.
User Experience and Privacy
The Chrome extension works on Chrome, Edge, Brave, and Arc, adding badges to posts on platforms like X, LinkedIn, Reddit, and Hacker News. The badges use a color-coded system: red for AI-written, amber for mixed content, and green for human-written text. Sniffy emphasizes privacy, stating that it does not store user text or results, and the extension contains no analytics or ads. Users can check up to 12,000 words at a time without signing up for an account.
Honest About Limitations
The creator openly admits that Sniffy is not yet as accurate as established competitors like Pangram and makes mistakes, particularly with short posts. The tool is designed to provide a hint rather than definitive proof, offering an estimate with a likely range for each detection. The developer plans to improve accuracy as more users contribute data, leveraging community feedback to refine the model's performance.
Key Takeaways
- Sniffy offers a completely free AI detection tool with no sign-up or paywall requirements.
- The extension integrates directly into social feeds, providing real-time AI detection badges.
- The underlying model is a fine-tuned RoBERTa, trained to detect text from ChatGPT, Claude, Gemini, and Llama.
- Privacy is prioritized, with no data storage and no analytics or ads in the extension.
- The tool is currently in an early stage, with acknowledged accuracy limitations compared to paid alternatives.
The Bottom Line
Sniffy is a promising, privacy-first entry into the crowded AI detection market, but its utility is currently capped by its acknowledged inaccuracy on short-form content. It serves best as a preliminary filter for developers and curious users rather than a definitive arbiter of truth. Until the model matures, treat every badge as a heuristic hint rather than hard evidence.