All legal documents

AI Transparency

Last updated: March 16, 2026

1. Our AI Approach

sh0 (operated by ZeroSuite, Inc.) integrates AI features to enhance the deployment experience. We believe in full transparency about how AI works on our platform, its capabilities, and its limitations.

2. AI Models Used

  • Primary Model: DeepSeek V3 (via OpenRouter)
  • Vision Model: GPT-4o Mini (for screenshot and UI analysis)
  • Provider Gateway: OpenRouter (openrouter.ai)
  • Architecture: Transformer-based large language models

3. How We Use AI

sh0 uses AI in the following ways:

3.1 Build Diagnostics

  • Analyzing build errors and suggesting fixes
  • Identifying missing dependencies or configuration issues
  • Providing context-aware troubleshooting

3.2 Resource Advisor

  • Analyzing container resource usage patterns
  • Recommending optimal CPU and memory allocation
  • Identifying resource bottlenecks

3.3 Chat-to-Deploy

  • Accepting natural language deployment requests
  • Translating intent into deployment configuration
  • Asking clarifying questions for better results

3.4 Code Health Check

  • Static analysis with 40+ rules (pure Rust, no AI required)
  • AI-enhanced analysis available for deeper insights
  • Security, performance, and best-practice checks

4. AI Capabilities

sh0 AI excels at:

  • Identifying common build and deployment errors
  • Suggesting configuration improvements
  • Explaining error messages in plain language
  • Recommending resource allocations based on usage patterns
  • Multi-language stack detection and support

5. AI Limitations

It is critical to understand what AI cannot do:

5.1 Knowledge Limitations

  • Knowledge Cutoff: Training data has a cutoff date and may not cover the latest versions
  • Real-time Data: Cannot access live metrics or current state
  • Specialized Knowledge: May lack expertise in niche or proprietary frameworks

5.2 Accuracy Limitations

  • Hallucinations: AI may confidently state incorrect information
  • False Fixes: Suggested fixes may not resolve the actual issue
  • Incomplete Analysis: May miss context from files it has not seen
  • Inconsistency: Results may vary between requests

5.3 Domain Limitations

  • Not DevOps Expertise: AI suggestions complement but do not replace professional knowledge
  • No Infrastructure Access: AI cannot directly modify your deployments
  • Security Limitations: AI should not be relied upon as the sole security auditor

6. Responsible AI Use

We recommend:

  • Always review and verify AI-generated suggestions before applying them
  • Test AI-recommended changes in a staging environment first
  • Be aware of hallucinations and validate important recommendations
  • Use AI as a tool to enhance human judgment, not replace it
  • Report inaccurate or problematic AI outputs to us

7. How We Improve AI

  • User Feedback: We collect feedback on AI quality
  • Model Updates: Regularly evaluate and upgrade to better models
  • Prompt Engineering: Continuously refine prompts for accuracy
  • Monitoring: Continuous quality assurance

8. Data Used for AI

  • Your Code: Sent to AI providers only during active analysis (not stored)
  • Build Logs: May be included for diagnostic context
  • No Training: Your data is not used to train AI models
  • Opt-out Available: You can disable all AI features

9. Transparency Reports

sh0 is committed to transparency:

  • We publish changelog entries for AI model changes
  • AI feature usage statistics are available in your dashboard
  • Self-hosted users have full control and visibility over AI usage

10. Contact for AI Questions

Questions about our AI approach?

11. Additional Resources