SlopTotal brings self-hosted AI text detection to local CPUs

Developer Pablo Caeg has released SlopTotal, an open-source project that functions as a local, self-hosted alternative to commercial AI content detectors. Published in late September 2026, the tool aggregates twenty-three different detection engines to analyze text, documents, and URLs entirely on the user’s own hardware. By running these models locally on CPU, it ensures that sensitive data never leaves the user's infrastructure while providing granular visibility into how each engine reaches its conclusion.

What happened

SlopTotal operates by parallelizing twenty-three independent detection methods, ranging from neural classifiers like DeBERTa and RoBERTa to statistical tests such as perplexity and burstiness analysis. Instead of returning a single black-box probability score, the system presents a calibrated ensemble verdict alongside individual scores from every engine. This approach allows users to inspect the specific contributions of linguistic heuristics, stock-phrase detectors, and deep learning models. The project includes a web interface and a JSON API, enabling integration into existing workflows or manual checks via a browser.

The tool also features a specialized "site check" capability designed to identify websites generated by AI app builders. It scans for specific digital fingerprints left by platforms like Lovable, v0, Bolt, Base44, Replit, and Same. Rather than guessing based on visual style or code quality, it looks for concrete markers such as specific JavaScript runtimes, meta tags, and hosting domains associated with these builders. This distinction helps separate AI-assisted coding from AI-generated deployment artifacts, providing evidence rather than just a probability percentage.

Key details

• Local execution: The software runs entirely on CPU, requiring no GPU, with a minimum of 4 GB RAM for the lite profile.
• Twenty-three engines: Combines neural classifiers, statistical methods, and linguistic heuristics for a multi-layered analysis.
• Transparent benchmarking: Accuracy metrics are published with raw per-sample scores, including tests against pre-1920 literature to measure false positives.
• Document support: Accepts direct text input, URLs, PDFs, Word documents, and Markdown files for scanning.
• Privacy focused: No data is sent to external APIs; reports are stored locally and deleted after thirty days by default.
• Docker ready: Can be deployed instantly using a single Docker command, with models caching automatically after the first run.

Background

AI detection has become a contentious field due to the high rate of false positives and the opacity of proprietary tools. Many commercial detectors operate as black boxes, offering a single confidence score without explaining the underlying logic. This lack of transparency makes it difficult for organizations to trust the results, especially when dealing with formal writing or older literary styles that may share structural similarities with machine-generated text. Furthermore, sending sensitive corporate documents to third-party detection services raises significant data privacy and security concerns.

SlopTotal addresses these issues by adopting an ensemble approach similar to VirusTotal, which aggregates multiple antivirus engines. By combining diverse detection strategies, it mitigates the weaknesses of individual models. For instance, while neural classifiers are powerful, they can be biased against certain writing styles. Linguistic heuristics, though weaker individually, provide independent signals that help break ties. The project explicitly publishes its failure cases and benchmarking methodology, allowing users to understand the limitations of the technology rather than relying on marketing claims.

Why it matters

For teams that manage their own software infrastructure, the ability to audit AI-generated content without exposing intellectual property to external vendors is crucial. Self-hosting SlopTotal ensures that internal documents, code reviews, and communications remain within the company’s security perimeter. This is particularly important for industries with strict compliance requirements where data leakage to third-party AI services is prohibited. The CPU-only requirement also lowers the barrier to entry, allowing deployment on standard servers or even developer laptops without specialized hardware.

The transparency of the scoring mechanism provides practical value for editorial and compliance workflows. When a piece of content is flagged, reviewers can see exactly which engines raised alarms and why. This granular insight helps distinguish between genuine AI generation and stylistic quirks in human writing. Additionally, the site-check feature offers a new way to assess vendor deliverables or internal projects, identifying whether a website was built using rapid AI prototyping tools, which may have implications for maintainability and security.

What you can do

• Deploy the tool locally using Docker to test its performance on your specific hardware configuration.
• Integrate the JSON API into your content management system to automate preliminary screening of submitted articles.
• Use the site-check endpoint to audit external vendor websites for signs of AI-generated boilerplate code.
• Review the published benchmark data to understand the tool’s false-positive rates before implementing it in critical workflows.
• Configure the retention policy to match your organization’s data governance standards, ensuring reports are deleted appropriately.
• Run the smoke test script after installation to verify that all twenty-three engines load and function correctly.