Devseis/endpoint-auditor-0.5b

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Oct 8, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Devseis/endpoint-auditor-0.5b is a 0.5 billion parameter language model developed by Devseis, fine-tuned from Qwen/Qwen2.5-0.5B-Instruct with a 32768 token context length. This model specializes in generating endpoint audit findings for ISO 27001, GDPR, and the EU AI Act from collected evidence. It is optimized for on-device deployment (e.g., WebLLM in the Devseis Endpoint Auditor desktop app) to ensure audit evidence remains local. The model excels at explaining findings, rating risks, and mapping them to compliance references, achieving 100% valid JSON and grounded responses in evaluation.

Loading preview...

Overview

Devseis Endpoint Auditor 0.5B (v0.6) is a compact language model developed by Devseis, specifically fine-tuned to automate the generation of endpoint audit findings. Built upon the Qwen/Qwen2.5-0.5B-Instruct base model, it processes collected evidence to produce detailed reports compliant with ISO 27001, GDPR, and the EU AI Act. A key differentiator is its design for on-device execution (e.g., via WebLLM in the Devseis Endpoint Auditor desktop app), ensuring that sensitive audit evidence never leaves the user's local machine.

This v0.6 iteration was specifically retrained to maintain performance when rounded to 4-bit precision for browser builds, addressing issues encountered with v0.5. It uses a 32768 token context length and is approximately 300 MB in 4-bit.

Key Capabilities

  • Automated Audit Reporting: Generates explanations for findings, risk ratings, and suggested fix/verification steps tailored to the device's OS.
  • Compliance Mapping: Maps findings to specific ISO 27001, GDPR, and EU AI Act references.
  • Executive Summaries: Produces concise executive summaries of audit results.
  • AI Tool Classification: Classifies AI tools under the EU AI Act framework.
  • Evidence Grounding: Every model answer is checked against provided evidence; if a check fails, the application uses a built-in template, ensuring factual accuracy.

Good For

  • On-device Compliance Auditing: Ideal for applications requiring local processing of sensitive audit data without cloud reliance.
  • Automating Security & Privacy Audits: Streamlining the generation of detailed reports for ISO 27001, GDPR, and EU AI Act compliance.
  • Developers Building Audit Tools: Provides a specialized LLM component for integrating automated audit finding generation into desktop or mobile applications.

Performance Highlights

Evaluation shows significant improvement over the base model, achieving 100% valid JSON and 100% grounded responses across ai_classification, finding, and summary tasks, and 92% for chat tasks. This ensures high reliability and faithfulness to the provided evidence.

Finetunes
4 models